Air-Gapped Deployment: Open-Source Embeddings, Ollama Config, MCP Compatibility (Wazuh 4.14.1)

[nikopuf]

Hi Team

I try to deploy Wazuh-OpenClaw-Autopilot in a fully air-gapped environment and need clarification on the following:

Memory / Embeddings (Offline Requirement)
Current configuration uses:

"memory": {
  "enabled": true,
  "search": {
    "provider": "openai",
    "model": "text-embedding-3-small",
    "hybrid": true
  }
}

Since we cannot use OpenAI in air-gapped mode:

1. Is there an officially supported open-source embedding solution?
2. Can Ollama (e.g., nomic-embed-text) be used for embeddings?
3. Is there support for local vector databases (Qdrant / Chroma / others)?
4. What is the recommended memory configuration for fully offline deployment?

Fully Air-Gapped Ollama Deployment:

We want to:

• Use Ollama as the only LLM provider
• Disable all cloud providers
• Run completely offline

Please confirm:

• Is Ollama officially supported as primary + fallback model?
• Is embedding via Ollama supported?
• Can you provide a fully offline openclaw.json example?

Wazuh Version Compatibility
We are currently using Wazuh App version: 4.14.1

Issue:
Most MCP endpoints are not responding.

Questions:

1. Is 4.14.1 officially supported?
2. What Wazuh version is recommended for stable MCP integration?
3. Are there known compatibility issues with 4.14.x?

Could you please consider adding an option in install.sh to allow installing only the MCP server or installing MCP together with Autopilot without Tailscale? Currently, even after setting AUTOPILOT_MODE=bootstrap and AUTOPILOT_REQUIRE_TAILSCALE=false in the .env file under the install directory, running ./install.sh still installs Tailscale. I would appreciate clarification on how to completely disable Tailscale installation, whether any modification in install.sh is required, and at what stage the openclaw.json file should be updated — after the default installation or by modifying install.sh before execution.

I already saw this https://github.com/gensecaihq/Wazuh-Openclaw-Autopilot/blob/main/docs/SCENARIOS.md but running ./install/install.sh --menu still give default installation.

Additionally, I would be grateful for detailed guidance on performing the installation using Ollama in a fully offline environment, including which Ollama models are recommended for Autopilot use cases and how to run the system locally without Tailscale for testing purposes. Finally, please let me know if there is an official discussion platform, such as Slack, Discord, or a forum, for Autopilot deployment and development support. Thank you for your support; your guidance on offline deployment, Ollama configuration, MCP compatibility, and installer customization would be greatly appreciated.

Thanks
nikopuf

[alokemajumder]

Hi,

Thanks for writing in . Allow me sometime to get back to you with all information. Some of the configs for example tailscale are deliberate so I need some more time to look into this.

Thanks

[alokemajumder]

Hi @nikopuf,

Thank you for the thorough report. We've done a deep analysis of every point you raised and want to give you a comprehensive response. This is a long reply — we wanted to address each question properly.

Please note that , we have just started building it and its in a very early stage, where extensive testing is still going on . So please proceed with caution .

---

1. Memory / Embeddings (Offline Requirement)

Current State

You're correct — the default openclaw.json hardcodes OpenAI for embeddings:

"memory": {
  "enabled": true,
  "search": {
    "provider": "openai",
    "model": "text-embedding-3-small",
    "hybrid": true
  }
}

This exists both in the reference template (openclaw/openclaw.json) and in the config that install.sh generates at ~/.openclaw/openclaw.json. Neither will work without OpenAI API access.

Answers to Your Questions

1. Is there an officially supported open-source embedding solution?

Not yet in the current release. The memory configuration delegates to OpenClaw's embedding abstraction layer. We are investigating adding Ollama embedding support (see below).

2. Can Ollama (e.g., nomic-embed-text) be used for embeddings?

This depends on OpenClaw's embedding provider abstraction. OpenClaw uses the configured provider to generate vectors for memory search. If OpenClaw supports "provider": "ollama" for the memory/search config, then models like nomic-embed-text or mxbai-embed-large should work. We are verifying this with the OpenClaw team and will update this issue once confirmed.

3. Is there support for local vector databases (Qdrant / Chroma / others)?

Not currently. OpenClaw uses its own built-in memory store for vector storage. There is no pluggable vector DB backend at this time.

4. What is the recommended memory configuration for fully offline deployment?

Disable memory entirely until offline embedding support is confirmed:

"memory": {
  "enabled": false
}

This is safe to do. Memory search is a convenience feature for cross-session context recall — it helps agents remember context from previous conversations. All core SOC functions (alert triage, correlation, investigation, response planning, reporting) work normally without it. You lose long-term agent memory across sessions, but each individual alert processing workflow is self-contained.

---

2. Fully Air-Gapped Ollama Deployment

Confirmation

• Is Ollama officially supported as primary + fallback model? Yes. Ollama is in the provider list, the model allowlist, and the fallback chain. It can be used as the sole provider.
• Is embedding via Ollama supported? Under investigation (see above). For now, disable memory.
• Can you provide a fully offline openclaw.json example? Yes — see below.

Complete Air-Gapped openclaw.json

Here is a complete, ready-to-use configuration for a fully air-gapped Ollama-only deployment. This replaces the entire ~/.openclaw/openclaw.json file:

{
  "_comment": "Wazuh OpenClaw Autopilot - Air-Gapped Configuration (Ollama Only)",
  "_security": [
    "Gateway binds to localhost only - never exposed to internet",
    "All cloud LLM providers disabled",
    "No external network calls required"
  ],

  "gateway": {
    "port": 18789,
    "bind": "loopback",
    "auth": {
      "token": "${OPENCLAW_TOKEN}"
    },
    "logging": {
      "redactSensitive": true
    },
    "security": {
      "cors": {
        "enabled": false
      },
      "rateLimit": {
        "enabled": true,
        "maxRequests": 100,
        "windowMs": 60000
      }
    }
  },

  "models": {
    "default": "ollama/llama3.3",

    "allowlist": [
      "ollama/llama3.3",
      "ollama/mistral",
      "ollama/codellama",
      "ollama/llama3.1",
      "ollama/qwen2.5"
    ],

    "fallbackChain": [
      "ollama/llama3.3",
      "ollama/mistral"
    ],

    "providers": {
      "ollama": {
        "enabled": true,
        "baseUrl": "http://127.0.0.1:11434/v1",
        "timeout": 300000
      },
      "anthropic": { "enabled": false },
      "openai": { "enabled": false },
      "google": { "enabled": false },
      "groq": { "enabled": false },
      "mistral": { "enabled": false },
      "xai": { "enabled": false },
      "openrouter": { "enabled": false },
      "together": { "enabled": false },
      "cerebras": { "enabled": false }
    }
  },

  "agents": {
    "defaults": {
      "workspace": "~/.openclaw/wazuh-autopilot/workspace",
      "agentDir": "~/.openclaw/wazuh-autopilot/agents",

      "model": {
        "primary": "ollama/llama3.3",
        "fallback": "ollama/mistral",
        "fast": "ollama/mistral"
      },

      "sandbox": {
        "mode": "all",
        "scope": "session",
        "workspaceAccess": "rw"
      },

      "tools": {
        "profile": "minimal",
        "deny": [
          "browser",
          "canvas",
          "nodes",
          "exec",
          "delete"
        ]
      },

      "heartbeat": {
        "every": "30m",
        "target": "last",
        "model": "ollama/mistral"
      },

      "memory": {
        "enabled": false
      }
    },

    "list": [
      {
        "id": "wazuh-triage",
        "default": true,
        "workspace": "~/.openclaw/wazuh-autopilot/agents/triage",
        "model": { "primary": "ollama/llama3.3" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        },
        "heartbeat": { "every": "10m" }
      },
      {
        "id": "wazuh-correlation",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/correlation",
        "model": { "primary": "ollama/llama3.3" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-investigation",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/investigation",
        "model": { "primary": "ollama/llama3.3" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-response-planner",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/response-planner",
        "model": { "primary": "ollama/llama3.3" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "write", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-policy-guard",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/policy-guard",
        "model": { "primary": "ollama/llama3.3" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-responder",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/responder",
        "model": { "primary": "ollama/llama3.3" },
        "tools": {
          "profile": "coding",
          "allow": ["read", "write", "exec", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["browser", "canvas", "delete"],
          "elevated": {
            "enabled": true,
            "allowFrom": ["slack"]
          }
        },
        "sandbox": {
          "mode": "all",
          "scope": "agent",
          "workspaceAccess": "rw"
        }
      },
      {
        "id": "wazuh-reporting",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/reporting",
        "model": { "primary": "ollama/mistral" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "write", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["exec", "delete", "browser", "canvas"]
        }
      }
    ]
  },

  "channels": {
    "slack": {
      "enabled": true,
      "dmPolicy": "allowlist",
      "allowFrom": [],
      "groupPolicy": "mention",
      "mentionGating": true,
      "pairing": {
        "enabled": true,
        "requireApproval": true
      }
    }
  },

  "tools": {
    "profile": "minimal",
    "allow": [
      "read",
      "sessions_list",
      "sessions_history"
    ],
    "deny": [
      "browser",
      "canvas",
      "nodes",
      "cron",
      "exec",
      "delete"
    ],
    "webSearch": {
      "enabled": false
    },
    "sandbox": {
      "allowlist": [
        "read",
        "sessions_list",
        "sessions_history",
        "sessions_send"
      ],
      "denylist": [
        "browser",
        "canvas",
        "exec",
        "delete"
      ]
    }
  },

  "automation": {
    "cron": [
      {"id": "hourly-snapshot", "schedule": "0 * * * *", "agentId": "wazuh-reporting"},
      {"id": "daily-digest", "schedule": "0 8 * * *", "agentId": "wazuh-reporting"},
      {"id": "shift-handoff", "schedule": "0 6,14,22 * * *", "agentId": "wazuh-reporting"},
      {"id": "weekly-summary", "schedule": "0 9 * * 1", "agentId": "wazuh-reporting"},
      {"id": "correlation-sweep", "schedule": "*/5 * * * *", "agentId": "wazuh-correlation"},
      {"id": "untriaged-sweep", "schedule": "*/10 * * * *", "agentId": "wazuh-triage"}
    ],
    "webhooks": [
      {"path": "/webhook/wazuh-alert", "target": "wazuh-triage"},
      {"path": "/webhook/case-created", "target": "wazuh-correlation"},
      {"path": "/webhook/investigation-request", "target": "wazuh-investigation"},
      {"path": "/webhook/plan-request", "target": "wazuh-response-planner"},
      {"path": "/webhook/policy-check", "target": "wazuh-policy-guard"},
      {"path": "/webhook/execute-action", "target": "wazuh-responder"}
    ]
  },

  "memory": {
    "enabled": false
  },

  "env": {
    "OPENCLAW_TOKEN": "${OPENCLAW_TOKEN}",
    "WAZUH_MCP_URL": "${MCP_URL}",
    "WAZUH_MCP_TOKEN": "${MCP_AUTH_TOKEN}"
  },

  "provider": {
    "type": "ollama",
    "baseUrl": "http://127.0.0.1:11434/v1",
    "timeout": 300000
  }
}

When to Apply This Config

Apply after the default installation completes. The installer generates a default ~/.openclaw/openclaw.json during the deploy_agents step. Replace that file with the air-gapped version above, then restart OpenClaw.

# 1. Run the installer first (see Section 4 for Tailscale workaround)
# 2. Replace the generated config:
cp your-airgapped-openclaw.json ~/.openclaw/openclaw.json
chmod 600 ~/.openclaw/openclaw.json
# 3. Restart OpenClaw
docker restart openclaw
# or if running as systemd:
sudo systemctl restart openclaw

Recommended Ollama Models for SOC Tasks

Model	Parameters	RAM Required	Best For
llama3.3	70B	~40 GB	Complex reasoning: triage analysis, investigation, response planning. Best overall SOC performance with local models.
llama3.1	8B/70B	~8 GB / ~40 GB	Good alternative to llama3.3. The 8B variant is lighter.
qwen2.5	7B/72B	~8 GB / ~40 GB	Strong multilingual support, good structured output.
mistral	7B	~8 GB	Lighter tasks: heartbeats, reporting summaries, quick status checks. Good for resource-constrained environments.
codellama	13B	~16 GB	Technical analysis, log parsing, rule interpretation.

Recommendation for air-gapped SOC: Use llama3.3 (70B) as the primary model for all agents and mistral (7B) for heartbeats and reporting. If your hardware can't support the 70B model, use llama3.1:8b as primary and mistral as fallback.

Pre-pull models before going air-gapped:

ollama pull llama3.3
ollama pull mistral
# Optional:
ollama pull codellama
ollama pull nomic-embed-text   # For when embedding support is confirmed

---

3. Wazuh 4.14.1 Compatibility

Version Support

We analyzed the Wazuh MCP Server (v4.0.6) in detail:

Wazuh Version	Support Level	Notes
4.14.x (including 4.14.1)	Fully Supported	Recommended. All 29 MCP tools work.
4.10.x – 4.13.x	Fully Supported	All features work
4.8.x – 4.9.x	Fully Supported	Minimum for vulnerability tools
4.0.0 – 4.7.x	Limited	No Wazuh Indexer — 3 vulnerability tools unavailable

Wazuh 4.14.1 is fully supported. The issue is not version compatibility.

Troubleshooting "Most MCP Endpoints Not Responding"

If most endpoints are not responding, the root cause is almost certainly one of the following:

Step 1 — Verify Wazuh API is reachable:

curl -k -u YOUR_WAZUH_USER:YOUR_PASSWORD https://YOUR_WAZUH_HOST:55000/

You should get a JSON response like:

{
  "data": {
    "title": "Wazuh Manager",
    "api_version": "4.14.1",
    ...
  }
}

If this fails, the problem is Wazuh API access (firewall, credentials, or the API service is not running), not the MCP server.

Step 2 — Check MCP server logs for errors:

# If running as Docker:
docker logs wazuh-mcp-server 2>&1 | tail -50

# If running as systemd:
journalctl -u wazuh-mcp-server -n 50

Look for authentication errors, connection timeouts, or SSL certificate issues.

Step 3 — SSL certificate issues:

Wazuh uses self-signed certificates by default. If the MCP server can't verify the cert, all requests will fail. Set these in the MCP server environment:

WAZUH_VERIFY_SSL=false
WAZUH_ALLOW_SELF_SIGNED=true

Step 4 — Wazuh API user permissions:

Ensure the API user has the correct roles. The MCP server needs access to alerts, agents, rules, and syscollector endpoints. The default wazuh-wui user should have sufficient permissions, but custom users may need additional roles assigned.

Step 5 — Wazuh Indexer (for vulnerability tools only):

3 of the 29 MCP tools (vulnerability scanning) require the Wazuh Indexer (OpenSearch-based, port 9200). If you don't have the Indexer running, those 3 tools will fail — but the other 26 tools should work normally. Check if the MCP server has these configured:

WAZUH_INDEXER_HOST=https://YOUR_INDEXER_HOST:9200
WAZUH_INDEXER_USER=admin
WAZUH_INDEXER_PASS=your_password

If you don't use the Indexer, that's fine — only the vulnerability summary/search tools are affected.

Step 6 — Network connectivity:

The MCP server needs to reach the Wazuh Manager API on port 55000 (HTTPS). In an air-gapped environment, ensure there's no firewall blocking traffic between the MCP server and Wazuh Manager.

# From the MCP server host:
curl -k https://YOUR_WAZUH_HOST:55000/ -o /dev/null -w "%{http_code}" -s
# Should return 401 (auth required) or 200 — anything except 000 (connection refused)

---

4. Installer: Tailscale, --mode, --menu

The Problem (Confirmed)

You've identified a genuine issue. After a thorough code review:

1. --mode and --menu flags are not implemented. The installer's main() function (line 1558) accepts "$@" but never parses arguments. Running ./install.sh --menu or ./install.sh --mode all-in-one is identical to running ./install.sh with no arguments — the flags are silently ignored.

2. Tailscale is unconditionally mandatory. install_tailscale is called at line 1573 with no conditions. If you decline Tailscale authentication, the installer exits (line 408). There is no --skip-tailscale option.

3. AUTOPILOT_MODE=bootstrap has no effect on the installer. The env var is only consumed by install/doctor.sh and the runtime service. The installer never reads it.

4. AUTOPILOT_REQUIRE_TAILSCALE=false has no effect on the installer. Same issue — only doctor.sh respects this variable.

The documentation in docs/SCENARIOS.md and docs/CLI_REFERENCE.md describes deployment modes and flags that exist as a design spec but are not yet implemented in the installer. This is something we need to fix.

Why Tailscale Is the Default

The Tailscale requirement is a deliberate security decision. In the production architecture:

• The OpenClaw Gateway binds to 127.0.0.1 (localhost only)
• The MCP Server binds to the Tailscale IP (never 0.0.0.0)
• The Runtime Service binds to 127.0.0.1
• No services are exposed to the public internet
• All inter-component traffic goes through Tailscale's encrypted, zero-trust network

This prevents accidental exposure of the MCP server (which has access to your Wazuh API) to untrusted networks. However, we acknowledge this should be skippable for air-gapped, evaluation, and testing environments.

Immediate Workaround

Until we ship the installer fix, here's how to install without Tailscale:

Option A — Comment out Tailscale in the installer:

# Edit install/install.sh before running it
# Find line 1573 and comment it out:

# install_tailscale    # Commented out for air-gapped install

You also need to ensure TAILSCALE_IP has a value, since later steps reference it. Either:

# Set it before running the installer:
export TAILSCALE_IP="127.0.0.1"
sudo -E ./install/install.sh

Or edit line 516 and line 931 in install.sh where TAILSCALE_IP is fetched — they already have || echo "127.0.0.1" fallbacks, so if Tailscale is not installed, the IP defaults to 127.0.0.1, which is correct for a single-server air-gapped setup.

Option B — Also skip the systemd dependency:

The generated systemd unit for the MCP server (wazuh-mcp-server.service) includes Requires=tailscaled.service (line 525). If Tailscale is not installed, the service won't start. After installation, edit the service file:

sudo nano /etc/systemd/system/wazuh-mcp-server.service
# Remove or comment out these lines:
#   After=... tailscaled.service
#   Requires=tailscaled.service

sudo systemctl daemon-reload
sudo systemctl restart wazuh-mcp-server

openclaw.json — When to Modify

The installer generates ~/.openclaw/openclaw.json during the deploy_agents step (line 597 in install.sh). The sequence is:

1. Installer runs through all steps including install_openclaw and deploy_agents
2. deploy_agents writes a default openclaw.json to ~/.openclaw/openclaw.json
3. After installation completes, you replace that file with your air-gapped version
4. Restart OpenClaw

You do not need to modify install.sh to change the openclaw.json — just overwrite the generated file after the installer finishes. The reference template at openclaw/openclaw.json in the repo is for documentation only; the installer uses a hardcoded heredoc (lines 597–837).

What We're Fixing

We're working on installer improvements that will include:

• A --skip-tailscale flag for air-gapped and testing environments
• Reading AUTOPILOT_MODE=bootstrap from the environment to auto-skip Tailscale
• Removing the Requires=tailscaled.service dependency when Tailscale is skipped
• Either implementing the --mode matrix or updating the documentation to match what's available

---

5. Community Platform

We've set up GitHub Discussions as the community platform:

• General Q&A and deployment help: Deployment Help & Troubleshooting #3
• Feature requests and ideas: Feature Requests & Integration Ideas #4
• Show and tell (share your deployment): Share Your Deployment #5

For ongoing discussion about your air-gapped deployment, feel free to continue in this issue or start a thread in Discussions.

---

Summary

Your Question	Answer	Status
Open-source embedding solution?	Not yet — disable memory for now	Investigating Ollama embeddings
Ollama for embeddings?	Under investigation with OpenClaw team	Will update this issue
Local vector DBs?	Not currently supported	OpenClaw uses built-in store
Recommended offline memory config?	memory.enabled: false	Safe, all core features work
Ollama as primary + fallback?	Yes, fully supported	See config above
Fully offline openclaw.json?	Provided above	Complete, ready-to-use
Wazuh 4.14.1 supported?	Yes, fully supported (4.8.0–4.14.3)	See troubleshooting steps
MCP endpoints not responding?	Likely auth/SSL/network issue	See 6-step troubleshooting
Install without Tailscale?	Workaround: comment out line 1573	Fix in progress
--mode/--menu not working?	Confirmed bug — flags not implemented	Fix in progress
AUTOPILOT_MODE=bootstrap?	Installer ignores it — only doctor.sh uses it	Fix in progress
When to modify openclaw.json?	After installation, replace ~/.openclaw/openclaw.json	Documented above
Recommended Ollama models?	llama3.3 (primary) + mistral (fast)	See model table
Community platform?	GitHub Discussions — now live	Links above

Thanks for the detailed report, @nikopuf. This has helped us identify real gaps in the installer and documentation. Please let us know how the air-gapped deployment goes and if you hit any other issues.

[nikopuf]

Hello @alokemajumder thanks for the detailed response by following above i got to this error when run this installation command

sudo ./install/install.sh --skip-tailscale

━━━ Installing Wazuh MCP Server (Localhost Only) ━━━

[INFO] Cloning Wazuh MCP Server from upstream...
Cloning into '/opt/wazuh-autopilot/wazuh-mcp-server'...
remote: Enumerating objects: 1921, done.
remote: Counting objects: 100% (245/245), done.
remote: Compressing objects: 100% (112/112), done.
remote: Total 1921 (delta 151), reused 159 (delta 129), pack-reused 1676 (from 2)
Receiving objects: 100% (1921/1921), 2.13 MiB | 32.00 KiB/s, done.
Resolving deltas: 100% (903/903), done.
[INFO] Installing dependencies...
npm error code ENOENT
npm error syscall open
npm error path /opt/wazuh-autopilot/wazuh-mcp-server/package.json
npm error errno -2
npm error enoent Could not read package.json: Error: ENOENT: no such file or directory, open '/opt/wazuh-autopilot/wazuh-mcp-server/package.json'
npm error enoent This is related to npm not being able to find a file.
npm error enoent
npm error A complete log of this run can be found in: /root/.npm/_logs/2026-02-17T12_28_33_604Z-debug-0.log

I check the directory and git repo the package.json is not available. could you guide in this above error.

Also could you please share the recommended MCP-server .env file which i can used and replace with default .env of MCP. It would be great. Thank you

Also the path for "modelPath": "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M.gguf" in memory is not set to this ~/.cache/node-llama-cpp/models/embeddinggemma-300M.gguf ?

nikopuf

[nikopuf]

Hi @alokemajumder i try the above guide with the openclaw-airgapped.json but it throw alot of error when it starting the openclaw gateway looks like the openclaw-airgapped.json is old version supported and not worked with openclaw version 2026.2.17 below are the status of openclaw gateway restart :

Invalid config at /root/.openclaw/openclaw.json:\n- models.providers.ollama.models: Invalid input: expected array, received undefined
- models.providers.ollama: Unrecognized keys: "enabled", "timeout"
- models: Unrecognized keys: "default", "allowlist", "fallbackChain"
- agents.defaults.model: Unrecognized keys: "fallback", "fast"
- agents.defaults: Unrecognized keys: "agentDir", "tools", "memory"
- agents.list.5.tools.elevated.allowFrom: Invalid input: expected record, received array
- tools.sandbox: Unrecognized keys: "allowlist", "denylist"
- tools: Unrecognized key: "webSearch"
- channels.slack.groupPolicy: Invalid option: expected one of "open"|"disabled"|"allowlist"
- channels.slack: Unrecognized keys: "mentionGating", "pairing"
- gateway: Unrecognized keys: "logging", "security"
- memory: Unrecognized keys: "enabled", "search", "compaction"
- <root>: Unrecognized keys: "_comment", "_security", "automation", "provider"

below is my openclaw.json also the installation dosn't read OPENCLAW_TOKEN , WAZUH_MCP_URL, WAZUH_MCP_TOKEN so i need to hard code it from /etc/wazuh-autopilot/.env

{
  "_comment": "Wazuh OpenClaw Autopilot - Air-Gapped Configuration (Ollama Only)",
  "_security": [
    "Gateway binds to localhost only - never exposed to internet",
    "All cloud LLM providers disabled",
    "Memory uses local GGUF embeddings (no cloud API needed)",
    "No external network calls required"
  ],

  "gateway": {
    "port": 18789,
    "bind": "loopback",
    "auth": {
      "token": "xxxxxxxx"
    },
    "logging": {
      "redactSensitive": true
    },
    "security": {
      "cors": {
        "enabled": false
      },
      "rateLimit": {
        "enabled": true,
        "maxRequests": 100,
        "windowMs": 60000
      }
    }
  },

  "models": {
    "default": "llama3.1:8b",

    "allowlist": [
      "mistral:latest",
      "codellama:latest",
      "ollama/llama3.1"
    ],

    "fallbackChain": [
      "llama3.1:8b",
      "mistral:latest"
    ],

    "providers": {
      "ollama": {
        "enabled": true,
        "baseUrl": "http://127.0.0.1:11434/v1",
        "timeout": 300000
      }
    }
  },

  "agents": {
    "defaults": {
      "workspace": "~/.openclaw/wazuh-autopilot/workspace",
      "agentDir": "~/.openclaw/wazuh-autopilot/agents",

      "model": {
        "primary": "llama3.1:8b",
        "fallback": "mistral:latest",
        "fast": "mistral:latest"
      },

      "sandbox": {
        "mode": "all",
        "scope": "session",
        "workspaceAccess": "rw"
      },

      "tools": {
        "profile": "minimal",
        "deny": [
          "browser",
          "canvas",
          "nodes",
          "exec",
          "delete"
        ]
      },

      "heartbeat": {
        "every": "30m",
        "target": "last",
        "model": "mistral:latest"
      },

      "memory": {
        "enabled": true,
        "search": {
          "provider": "local",
          "local": {
            "modelPath": "/root/.cache/embeddinggemma-300M.gguf"
          },
          "hybrid": true,
          "bm25Weight": 0.4,
          "vectorWeight": 0.6
        }
      }
    },

    "list": [
      {
        "id": "wazuh-triage",
        "default": true,
        "workspace": "~/.openclaw/wazuh-autopilot/agents/triage",
        "model": { "primary": "llama3.1:8b" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        },
        "heartbeat": { "every": "10m" }
      },
      {
        "id": "wazuh-correlation",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/correlation",
        "model": { "primary": "llama3.1:8b" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-investigation",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/investigation",
        "model": { "primary": "llama3.1:8b" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-response-planner",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/response-planner",
        "model": { "primary": "llama3.1:8b" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "write", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-policy-guard",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/policy-guard",
        "model": { "primary": "llama3.1:8b" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser", "canvas"]
        }
      },
      {
        "id": "wazuh-responder",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/responder",
        "model": { "primary": "llama3.1:8b" },
        "tools": {
          "profile": "coding",
          "allow": ["read", "write", "exec", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["browser", "canvas", "delete"],
          "elevated": {
            "enabled": true,
            "allowFrom": ["slack"]
          }
        },
        "sandbox": {
          "mode": "all",
          "scope": "agent",
          "workspaceAccess": "rw"
        }
      },
      {
        "id": "wazuh-reporting",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/reporting",
        "model": { "primary": "mistral:latest" },
        "tools": {
          "profile": "minimal",
          "allow": ["read", "write", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["exec", "delete", "browser", "canvas"]
        }
      }
    ]
  },

  "channels": {
    "slack": {
      "enabled": true,
      "dmPolicy": "allowlist",
      "allowFrom": [],
      "groupPolicy": "mention",
      "mentionGating": true,
      "pairing": {
        "enabled": true,
        "requireApproval": true
      }
    }
  },

  "tools": {
    "profile": "minimal",
    "allow": [
      "read",
      "sessions_list",
      "sessions_history"
    ],
    "deny": [
      "browser",
      "canvas",
      "nodes",
      "cron",
      "exec",
      "delete"
    ],
    "webSearch": {
      "enabled": false
    },
    "sandbox": {
      "allowlist": [
        "read",
        "sessions_list",
        "sessions_history",
        "sessions_send"
      ],
      "denylist": [
        "browser",
        "canvas",
        "exec",
        "delete"
      ]
    }
  },

  "automation": {
    "cron": [
      {"id": "hourly-snapshot", "schedule": "0 * * * *", "agentId": "wazuh-reporting"},
      {"id": "daily-digest", "schedule": "0 8 * * *", "agentId": "wazuh-reporting"},
      {"id": "shift-handoff", "schedule": "0 6,14,22 * * *", "agentId": "wazuh-reporting"},
      {"id": "weekly-summary", "schedule": "0 9 * * 1", "agentId": "wazuh-reporting"},
      {"id": "correlation-sweep", "schedule": "*/5 * * * *", "agentId": "wazuh-correlation"},
      {"id": "untriaged-sweep", "schedule": "*/10 * * * *", "agentId": "wazuh-triage"}
    ],
    "webhooks": [
      {"path": "/webhook/wazuh-alert", "target": "wazuh-triage"},
      {"path": "/webhook/case-created", "target": "wazuh-correlation"},
      {"path": "/webhook/investigation-request", "target": "wazuh-investigation"},
      {"path": "/webhook/plan-request", "target": "wazuh-response-planner"},
      {"path": "/webhook/policy-check", "target": "wazuh-policy-guard"},
      {"path": "/webhook/execute-action", "target": "wazuh-responder"}
    ]
  },

  "memory": {
    "enabled": true,
    "search": {
      "provider": "local",
      "local": {
        "modelPath": "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M.gguf"
      },
      "hybrid": true,
      "bm25Weight": 0.4,
      "vectorWeight": 0.6
    },
    "compaction": {
      "enabled": true,
      "schedule": "daily"
    }
  },

  "env": {
    "OPENCLAW_TOKEN": "xxxxxxxxx",
    "WAZUH_MCP_URL": "http://127.0.0.1:3000",
    "WAZUH_MCP_TOKEN": "xxxxxxxxxx"
  },

  "provider": {
    "type": "ollama",
    "baseUrl": "http://127.0.0.1:11434/v1",
    "timeout": 300000
  }
}

or is there is any specific version of openclaw need to used for above config. Also the install.sh failed on run_security_audit if no slack and claude key setup so i need comment it out for work. Can you please provide the openclaw.json with ollama with openclaw version 2026.2.17.

Thanks
nikopuf

[alokemajumder]

Hi ,

Will look into this. Sorry for the trouble. Btw have you successfully installed and connected Mcp server with your wazuh?

[nikopuf]

I just installed the MCP can you tell me how should i validate all the 29 tools. below is the .env of my MCP server

# WAZUH CONNECTION (used by Wazuh MCP Server)
WAZUH_HOST=127.0.0.1
WAZUH_PORT=55000
WAZUH_USER=wazuh-wui
WAZUH_PASS=xxxxx
WAZUH_VERIFY_SSL=false
WAZUH_ALLOW_SELF_SIGNED=true

# MCP SERVER AUTHENTICATION
AUTH_MODE=bearer
MCP_API_KEY=6918fbc2f13d81f8dc0f533cc331d0c4b570349e055e8700e8d67f5b795f921a

# MCP SERVER NETWORK
MCP_HOST=127.0.0.1
MCP_PORT=3000
MCP_URL=http://127.0.0.1:3000
AUTOPILOT_MCP_AUTH=6918fbc2f13d81f8dc0f533cc331d0c4b570349e055e8700e8d67f5b795f921a

# OPENCLAW GATEWAY (Localhost Only)
OPENCLAW_HOST=127.0.0.1
OPENCLAW_PORT=18789
OPENCLAW_TOKEN=1fa6cfd93eb0406667bd27747e8f42ae62b215251eb030c189d48ad3b1181f75

# AI PROVIDER
ANTHROPIC_API_KEY=sk-ant-11223344556677889900oopp

# RUNTIME SERVICE (Localhost Only)
AUTOPILOT_HOST=127.0.0.1
AUTOPILOT_PORT=9090
AUTOPILOT_DATA_DIR=/var/lib/wazuh-autopilot
AUTOPILOT_LOG_DIR=/var/log/wazuh-autopilot
AUTOPILOT_CONFIG_DIR=/etc/wazuh-autopilot

# APPROVAL SYSTEM
AUTOPILOT_TOKEN_SECRET=a68776f307576b0c78e7568aaccb5fd1ac0a904449b1f3a80c5efd4914460472
AUTOPILOT_TOKEN_TTL_MINUTES=60

# PAIRING MODE
AUTOPILOT_PAIRING_CODE=b8ab03788653573d09bc938f2d404088
AUTOPILOT_PAIRING_ENABLED=true

# RESPONDER AGENT - TWO-TIER HUMAN APPROVAL
# SAFETY: Disabled by default. Every action requires human Approve + Execute.
AUTOPILOT_RESPONDER_ENABLED=true

# SLACK INTEGRATION
SLACK_APP_TOKEN=
SLACK_BOT_TOKEN=
SLACK_ALERTS_CHANNEL=
SLACK_APPROVALS_CHANNEL=

# TAILSCALE
TAILSCALE_IP=127.0.0.1

For the moment i get this service wazuh-mcp-server status

service wazuh-mcp-server status
● wazuh-mcp-server.service - Wazuh MCP Server
     Loaded: loaded (/etc/systemd/system/wazuh-mcp-server.service; enabled; preset: enabled)
     Active: active (running) since Thu 2026-02-19 10:00:41 PKT; 1h 16min ago
       Docs: https://github.com/gensecaihq/Wazuh-MCP-Server
   Main PID: 3842667 (python)
      Tasks: 6 (limit: 76600)
     Memory: 60.6M (peak: 61.9M)
        CPU: 6.562s
     CGroup: /system.slice/wazuh-mcp-server.service
             └─3842667 /opt/wazuh-autopilot/wazuh-mcp-server/.venv/bin/python -m wazuh_mcp_server

Feb 19 10:00:42 virt-5378 python[3842667]: INFO:     Uvicorn running on http://127.0.0.1:3000 (Press CTRL+C to quit)
Feb 19 10:02:17 virt-5378 python[3842667]: 2026-02-19 10:02:17,220 - httpx - INFO - HTTP Request: POST https://127.0.0.1:55000/security/user/authenticate "HTTP/1.1 200 OK"
Feb 19 10:02:17 virt-5378 python[3842667]: 2026-02-19 10:02:17,222 - wazuh_mcp_server.api.wazuh_client - INFO - Authenticated with Wazuh server at 127.0.0.1
Feb 19 10:02:17 virt-5378 python[3842667]: 2026-02-19 10:02:17,224 - wazuh_mcp_server.api.wazuh_indexer - INFO - WazuhIndexerClient initialized for 127.0.0.1:9200
Feb 19 10:02:17 virt-5378 python[3842667]: 2026-02-19 10:02:17,225 - wazuh_mcp_server.api.wazuh_client - INFO - Wazuh Indexer client initialized successfully
Feb 19 10:02:24 virt-5378 python[3842667]: 2026-02-19 10:02:24,902 - httpx - INFO - HTTP Request: GET https://127.0.0.1:55000/ "HTTP/1.1 200 OK"
Feb 19 10:02:24 virt-5378 python[3842667]: 2026-02-19 10:02:24,925 - httpx - INFO - HTTP Request: GET https://127.0.0.1:9200/_cluster/health "HTTP/1.1 200 OK"
Feb 19 10:02:24 virt-5378 python[3842667]: INFO:     127.0.0.1:34300 - "GET /health HTTP/1.1" 200 OK
Feb 19 10:02:37 virt-5378 python[3842667]: 2026-02-19 10:02:37,316 - httpx - INFO - HTTP Request: GET https://127.0.0.1:9200/_cluster/health "HTTP/1.1 200 OK"
Feb 19 10:02:37 virt-5378 python[3842667]: INFO:     127.0.0.1:39710 - "GET /health HTTP/1.1" 200 OK

curl http://localhost:3000/health

{"status":"healthy","timestamp":"2026-02-19T06:18:35.182240+00:00","version":"4.0.6","mcp_protocol_version":"2025-11-25","supported_protocol_versions":["2025-11-25","2025-06-18","2025-03-26","2024-11-05"],"transport":{"streamable_http":"enabled","legacy_sse":"enabled"},"authentication":{"mode":"bearer","bearer_enabled":true,"oauth_enabled":false,"authless":false},"services":{"wazuh_manager":"healthy","wazuh_indexer":"healthy","mcp":"healthy"},"vulnerability_tools":{"available":true,"note":"Wazuh Indexer configured"},"metrics":{"active_sessions":0,"total_sessions":0},"endpoints":{"recommended":"/mcp (Streamable HTTP - 2025-11-25)","legacy":"/sse (SSE only)","authentication":"/auth/token","monitoring":["/health","/metrics"]}}

can you please check if anything is missed or required ? Appreciate your response.

Thanks

[nikopuf]

Hi @alokemajumder any quick way to fix above openclaw issue.

Thanks
@nikopuf

[nikopuf]

Hi @alokemajumder can you please share your testing openclaw version for current mentioned openclaw-airgapped.json.

Thanks
@nikopuf

[alokemajumder]

Hi @nikopuf,

Thanks for the detailed feedback — the OpenClaw config schema issue has been identified and fixed.

---

OpenClaw Configuration Fix

The openclaw.json and openclaw-airgapped.json have been rewritten to be fully compatible with OpenClaw v2026.2.17. The previous configs had several keys that don't exist in the current OpenClaw schema, which is why you were getting validation errors.

To update, pull the latest from the repo and replace your config:

git pull

# For cloud LLM providers:
cp openclaw/openclaw.json ~/.openclaw/openclaw.json

# For air-gapped / Ollama-only:
cp openclaw/openclaw-airgapped.json ~/.openclaw/openclaw.json

# Verify config is valid:
openclaw doctor --fix

The air-gapped config now properly registers Ollama as a custom model provider, which was the main schema issue for your setup.

---

MCP Tool Validation

Your MCP server is running healthy — both Wazuh Manager and Indexer are connected, which means all 29 tools should be available. To validate:

# List all available tools
curl -X POST http://localhost:3000/mcp \
  -H "Authorization: Bearer YOUR_MCP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "tools/list", "id": 1}'

# Test alert retrieval
curl -X POST http://localhost:3000/mcp \
  -H "Authorization: Bearer YOUR_MCP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "tools/call", "params": {"name": "get_wazuh_alerts", "arguments": {"limit": 5}}, "id": 2}'

Replace YOUR_MCP_API_KEY with your actual MCP_API_KEY value from your .env.

---

Security Audit / Validation Fix

The installer validation has been updated so that missing cloud LLM API keys are treated as a warning (not a failure) when Ollama is available locally. Air-gapped deployments no longer need cloud provider keys to pass validation.

---

Your .env

Your .env looks correct. One note — you have AUTOPILOT_RESPONDER_ENABLED=true, so make sure you've reviewed the policy files under /etc/wazuh-autopilot/policies/ before triggering any response actions.

Let me know if the updated configs resolve the OpenClaw validation errors.

[nikopuf]

Hi @alokemajumder thanks for detailed response i get this error

openclaw doctor --fix
Invalid config at /root/.openclaw/openclaw.json:\n- models.providers.ollama.models.0.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.1.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.2.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.3.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.4.name: Invalid input: expected string, received undefined
- bindings.0.match.peer.id: Invalid input: expected string, received undefined
- bindings.1.match.peer.kind: Invalid input

🦞 OpenClaw 2026.2.17 (4134875) — I don't judge, but your missing API keys are absolutely judging you.

▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄
██░▄▄▄░██░▄▄░██░▄▄▄██░▀██░██░▄▄▀██░████░▄▄▀██░███░██
██░███░██░▀▀░██░▄▄▄██░█░█░██░█████░████░▀▀░██░█░█░██
██░▀▀▀░██░█████░▀▀▀██░██▄░██░▀▀▄██░▀▀░█░██░██▄▀▄▀▄██
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
                  🦞 OPENCLAW 🦞

┌  OpenClaw doctor
│
◇  Config ───────────────────────────────────────────────────╮
│                                                            │
│  Config invalid; doctor will run with best-effort config.  │
│                                                            │
├────────────────────────────────────────────────────────────╯
│
◇  Gateway ──────────────────────────────────────────────────────────╮
│                                                                    │
│  gateway.mode is unset; gateway start will be blocked.             │
│  Fix: run openclaw configure and set Gateway mode (local/remote).  │
│  Or set directly: openclaw config set gateway.mode local           │
│                                                                    │
├────────────────────────────────────────────────────────────────────╯
│
◇  Sandbox ───────────────────────────────────────────────────────────────────╮
│                                                                             │
│  Sandbox base image missing: openclaw-sandbox:bookworm-slim. Build it with  │
│  scripts/sandbox-setup.sh.                                                  │
│                                                                             │
├─────────────────────────────────────────────────────────────────────────────╯
│
◇  Sandbox ───────────────────────────────────────────────────────────────╮
│                                                                         │
│  Unable to locate scripts/sandbox-setup.sh. Run it from the repo root.  │
│                                                                         │
├─────────────────────────────────────────────────────────────────────────╯
│
◇  Security ─────────────────────────────────╮
│                                            │
│  - No channel security warnings detected.  │
│  - Run: openclaw security audit --deep     │
│                                            │
├────────────────────────────────────────────╯
│
◇  Skills status ────────────╮
│                            │
│  Eligible: 4               │
│  Missing requirements: 46  │
│  Blocked by allowlist: 0   │
│                            │
├────────────────────────────╯
Invalid config at /root/.openclaw/openclaw.json:\n- models.providers.ollama.models.0.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.1.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.2.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.3.name: Invalid input: expected string, received undefined
- models.providers.ollama.models.4.name: Invalid input: expected string, received undefined
- bindings.0.match.peer.id: Invalid input: expected string, received undefined
- bindings.1.match.peer.kind: Invalid input
│
◇  Plugins ──────╮
│                │
│  Loaded: 4     │
│  Disabled: 32  │
│  Errors: 0     │
│                │
├────────────────╯
│
◇  Memory search ────────────────────────────────────────────────────────────────────────────╮
│                                                                                            │
│  Memory search provider is set to "local" but no local model file was found.               │
│                                                                                            │
│  Fix (pick one):                                                                           │
│  - Install node-llama-cpp and set a local model path in config                             │
│  - Switch to a remote provider: openclaw config set agents.defaults.memorySearch.provider  │
│    openai                                                                                  │
│                                                                                            │
│  Verify: openclaw memory status --deep                                                     │
│                                                                                            │
├────────────────────────────────────────────────────────────────────────────────────────────╯
│
◇
│
◇  Gateway ──────────────╮
│                        │
│  Gateway not running.  │
│                        │
├────────────────────────╯
│
◇  Gateway connection ────────────────────╮
│                                         │
│  Gateway target: ws://127.0.0.1:18789   │
│  Source: local loopback                 │
│  Config: /root/.openclaw/openclaw.json  │
│  Bind: loopback                         │
│                                         │
├─────────────────────────────────────────╯
│
◇  Gateway ─────────────────────────────────────────────────────────────────────────╮
│                                                                                   │
│  Runtime: running (pid 572833, state active, sub running, last exit 0, reason 0)  │
│                                                                                   │
├───────────────────────────────────────────────────────────────────────────────────╯
Restarted systemd service: openclaw-gateway.service
│
◇
│
◇  Gateway ──────────────╮
│                        │
│  Gateway not running.  │
│                        │
├────────────────────────╯
│
◇  Gateway connection ────────────────────╮
│                                         │
│  Gateway target: ws://127.0.0.1:18789   │
│  Source: local loopback                 │
│  Config: /root/.openclaw/openclaw.json  │
│  Bind: loopback                         │
│                                         │
├─────────────────────────────────────────╯
Error: Config validation failed: models.providers.ollama.models.0.name: Invalid input: expected string, received undefined

this is my openclaw.json

{
  // =============================================================================
  // WAZUH OPENCLAW AUTOPILOT — Air-Gapped Configuration (Ollama Only)
  // =============================================================================
  // JSON5 format: comments and trailing commas are allowed.
  //
  // This configuration runs entirely on local Ollama models with NO external
  // network calls. Suitable for air-gapped, classified, or offline environments.
  //
  // Prerequisites:
  //   1. Install Ollama: https://ollama.com/download
  //   2. Pull models: ollama pull llama3.3 && ollama pull mistral
  //   3. Ensure Ollama is running: ollama serve (default http://127.0.0.1:11434)
  //
  // To use different models, update the models.providers.ollama.models array
  // and the agents.defaults.model.primary/fallbacks to match.
  // =============================================================================

  // ---------------------------------------------------------------------------
  // GATEWAY
  // ---------------------------------------------------------------------------
  "gateway": {
    "port": 18789,
    "bind": "loopback",
    "auth": {
      "mode": "token",
      "token": "5645bd9a521d0152fb0f4564b09fef99263f3a08a3ffea5c210ad83e42d046a8"
    }
  },

  // ---------------------------------------------------------------------------
  // LOGGING
  // ---------------------------------------------------------------------------
  "logging": {
    "redactSensitive": "tools"
  },

  // ---------------------------------------------------------------------------
  // CUSTOM MODEL PROVIDERS — Ollama via OpenAI-compatible API
  // ---------------------------------------------------------------------------
  // Ollama is not a built-in provider; register it here as a custom provider.
  // The baseUrl points to Ollama's OpenAI-compatible endpoint.
  // ---------------------------------------------------------------------------
  "models": {
    "providers": {
      "ollama": {
        "baseUrl": "http://127.0.0.1:11434/v1",
        "apiKey": "ollama",
        "api": "openai-completions",
        "models": [
          { "id": "llama3.3", "contextWindow": 131072 },
          { "id": "mistral", "contextWindow": 32768 },
          { "id": "codellama", "contextWindow": 16384 },
          { "id": "llama3.1", "contextWindow": 131072 },
          { "id": "qwen2.5", "contextWindow": 131072 }
        ]
      }
    }
  },

  // ---------------------------------------------------------------------------
  // AGENT CONFIGURATION
  // ---------------------------------------------------------------------------
  "agents": {
    "defaults": {
      "workspace": "~/.openclaw/wazuh-autopilot/workspace",

      "model": {
        "primary": "ollama/llama3.3",
        "fallbacks": ["ollama/mistral"]
      },

      "models": {
        "ollama/llama3.3": { "alias": "llama" },
        "ollama/mistral": { "alias": "mistral" }
      },

      "sandbox": {
        "mode": "all",
        "scope": "session"
      },

      "heartbeat": {
        "every": "30m",
        "model": "ollama/mistral"
      },

      // Local embeddings — no cloud API needed
      "memorySearch": {
        "sources": ["memory", "sessions"],
        "provider": "local",
        "model": "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M.gguf"
      },

      "maxConcurrent": 2,
      "timeoutSeconds": 900
    },

    "list": [
      {
        "id": "wazuh-triage",
        "default": true,
        "workspace": "~/.openclaw/wazuh-autopilot/agents/triage",
        "agentDir": "~/.openclaw/wazuh-autopilot/agents/triage",
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser"]
        },
        "heartbeat": { "every": "10m" }
      },
      {
        "id": "wazuh-correlation",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/correlation",
        "agentDir": "~/.openclaw/wazuh-autopilot/agents/correlation",
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser"]
        },
        "heartbeat": { "every": "5m" }
      },
      {
        "id": "wazuh-investigation",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/investigation",
        "agentDir": "~/.openclaw/wazuh-autopilot/agents/investigation",
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser"]
        }
      },
      {
        "id": "wazuh-response-planner",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/response-planner",
        "agentDir": "~/.openclaw/wazuh-autopilot/agents/response-planner",
        "tools": {
          "profile": "minimal",
          "allow": ["read", "write", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["exec", "delete", "browser"]
        }
      },
      {
        "id": "wazuh-policy-guard",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/policy-guard",
        "agentDir": "~/.openclaw/wazuh-autopilot/agents/policy-guard",
        "tools": {
          "profile": "minimal",
          "allow": ["read", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["write", "edit", "exec", "delete", "browser"]
        }
      },
      {
        "id": "wazuh-responder",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/responder",
        "agentDir": "~/.openclaw/wazuh-autopilot/agents/responder",
        "model": {
          "primary": "ollama/llama3.3",
          "fallbacks": ["ollama/mistral"]
        },
        "tools": {
          "profile": "coding",
          "allow": ["read", "write", "exec", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["browser", "delete"],
          "elevated": { "enabled": true }
        },
        "sandbox": {
          "mode": "all",
          "scope": "agent"
        }
      },
      {
        "id": "wazuh-reporting",
        "workspace": "~/.openclaw/wazuh-autopilot/agents/reporting",
        "agentDir": "~/.openclaw/wazuh-autopilot/agents/reporting",
        "model": "ollama/mistral",
        "tools": {
          "profile": "minimal",
          "allow": ["read", "write", "sessions_list", "sessions_history", "sessions_send"],
          "deny": ["exec", "delete", "browser"]
        }
      }
    ]
  },

  // ---------------------------------------------------------------------------
  // CHANNEL CONFIGURATION
  // ---------------------------------------------------------------------------
//  "channels": {
//    "slack": {
//      "enabled": false,
//      "botToken": "${SLACK_BOT_TOKEN}",
//      "appToken": "${SLACK_APP_TOKEN}",
//     "dmPolicy": "allowlist",
//      "allowFrom": [],
//      "groupPolicy": "allowlist"
//    }
//  },

  // ---------------------------------------------------------------------------
  // MULTI-AGENT ROUTING
  // ---------------------------------------------------------------------------
  "bindings": [
    {
      "agentId": "wazuh-triage",
      "match": {
        "channel": "slack",
        "peer": { "kind": "group" }
      }
    },
    {
      "agentId": "wazuh-responder",
      "match": {
        "channel": "slack",
        "peer": { "id": "approvals" }
      }
    }
  ],

  // ---------------------------------------------------------------------------
  // GLOBAL TOOL POLICY
  // ---------------------------------------------------------------------------
  "tools": {
    "profile": "minimal",
    "allow": ["read", "sessions_list", "sessions_history"],
    "deny": ["browser", "canvas"],
    "web": {
      "search": { "enabled": false },
      "fetch": { "enabled": false }
    }
  },

  // ---------------------------------------------------------------------------
  // CRON
  // ---------------------------------------------------------------------------
  "cron": {
    "enabled": true,
    "maxConcurrentRuns": 2
  },

  // ---------------------------------------------------------------------------
  // HOOKS — Webhook endpoints for Wazuh alert routing
  // ---------------------------------------------------------------------------
  "hooks": {
    "enabled": true,
    "token": "5645bd9a521d0152fb0f4564b09fef99263f3a08a3ffea5c210ad83e42d046a8",
    "mappings": [
      {
        "match": { "path": "/webhook/wazuh-alert" },
        "action": "agent",
        "agentId": "wazuh-triage"
      },
      {
        "match": { "path": "/webhook/case-created" },
        "action": "agent",
        "agentId": "wazuh-correlation"
      },
      {
        "match": { "path": "/webhook/investigation-request" },
        "action": "agent",
        "agentId": "wazuh-investigation"
      },
      {
        "match": { "path": "/webhook/plan-request" },
        "action": "agent",
        "agentId": "wazuh-response-planner"
      },
      {
        "match": { "path": "/webhook/policy-check" },
        "action": "agent",
        "agentId": "wazuh-policy-guard"
      },
      {
        "match": { "path": "/webhook/execute-action" },
        "action": "agent",
        "agentId": "wazuh-responder"
      }
    ]
  },

  // ---------------------------------------------------------------------------
  // ENVIRONMENT VARIABLES
  // ---------------------------------------------------------------------------
  "env": {
    "WAZUH_MCP_URL": "http://127.0.0.1:3000",
    "WAZUH_MCP_TOKEN": "dcf264d13b0107d835e077e029beb495dfeb692bbdb7331f8d5fa99781d55d76"
  }
}

this is my /etc/wazuh-autopilot.env

WAZUH_HOST=127.0.0.1
WAZUH_PORT=55000
WAZUH_USER=wazuh-wui
WAZUH_PASS=MyS3cr37P450r.*-
WAZUH_VERIFY_SSL=false
WAZUH_ALLOW_SELF_SIGNED=true

WAZUH_INDEXER_HOST=127.0.0.1
WAZUH_INDEXER_PORT=9200
WAZUH_INDEXER_USER=admin
WAZUH_INDEXER_PASS=SecretPassword

# MCP SERVER AUTHENTICATION
AUTH_MODE=bearer
MCP_API_KEY=dcf264d13b0107d835e077e029beb495dfeb692bbdb7331f8d5fa99781d55d76

# MCP SERVER NETWORK
MCP_HOST=127.0.0.1
MCP_PORT=3000
MCP_URL=http://127.0.0.1:3000
AUTOPILOT_MCP_AUTH=dcf264d13b0107d835e077e029beb495dfeb692bbdb7331f8d5fa99781d55d76

# OPENCLAW GATEWAY (Localhost Only)
OPENCLAW_HOST=127.0.0.1
OPENCLAW_PORT=18789
OPENCLAW_TOKEN=5645bd9a521d0152fb0f4564b09fef99263f3a08a3ffea5c210ad83e42d046a8

# AI PROVIDER
ANTHROPIC_API_KEY=

# RUNTIME SERVICE (Localhost Only)
AUTOPILOT_HOST=127.0.0.1
AUTOPILOT_PORT=9090
AUTOPILOT_DATA_DIR=/var/lib/wazuh-autopilot
AUTOPILOT_LOG_DIR=/var/log/wazuh-autopilot
AUTOPILOT_CONFIG_DIR=/etc/wazuh-autopilot

# APPROVAL SYSTEM
AUTOPILOT_TOKEN_SECRET=7b344572f5641907203c3a86fc1561d83b94b21d193c68c375d22545350b320f
AUTOPILOT_TOKEN_TTL_MINUTES=60

# PAIRING MODE
AUTOPILOT_PAIRING_CODE=533073947d721ae0ec3f556c3fc34830
AUTOPILOT_PAIRING_ENABLED=true

# RESPONDER AGENT - TWO-TIER HUMAN APPROVAL
# SAFETY: Disabled by default. Every action requires human Approve + Execute.
AUTOPILOT_RESPONDER_ENABLED=false

# SLACK INTEGRATION
SLACK_APP_TOKEN=
SLACK_BOT_TOKEN=
SLACK_ALERTS_CHANNEL=
SLACK_APPROVALS_CHANNEL=

# TAILSCALE
TAILSCALE_IP=127.0.0.1

also on checking wazuh-mcp i get this

curl -X POST http://localhost:3000/mcp \
  -H "Authorization: Bearer dcf264d13b0107d835e077e029beb495dfeb692bbdb7331f8d5fa99781d55d76" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "tools/list", "id": 1}'

{"detail":"Invalid or expired token"}

Also can we test the whole workflow without slack for the testing if yes can you guide

can you please guide. Thanks

@nikopuf

[nikopuf]

Hi @alokemajumder so i comment out this part

  // ---------------------------------------------------------------------------
  // CHANNEL CONFIGURATION
  // ---------------------------------------------------------------------------
//  "channels": {
//    "slack": {
//      "enabled": false,
//      "botToken": "${SLACK_BOT_TOKEN}",
//      "appToken": "${SLACK_APP_TOKEN}",
//     "dmPolicy": "allowlist",
//      "allowFrom": [],
//      "groupPolicy": "allowlist"
//    }
//  },

  // ---------------------------------------------------------------------------
  // MULTI-AGENT ROUTING
  // ---------------------------------------------------------------------------
//  "bindings": [
//    {
//      "agentId": "wazuh-triage",
//      "match": {
//        "channel": "slack",
//        "peer": { "kind": "group" }
//      }
//    },
//    {
//      "agentId": "wazuh-responder",
//      "match": {
//        "channel": "slack",
//        "peer": { "id": "approvals" }
//      }
//    }
//  ],

then i run this openclaw status --all i get this.

openclaw status --all

🦞 OpenClaw 2026.2.17 (4134875) — OpenAI-compatible, not OpenAI-dependent.

│
OpenClaw status --all

Overview
┌─────────────────┬───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Item            │ Value                                                                                                                             │
├─────────────────┼───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Version         │ 2026.2.17                                                                                                                         │
│ OS              │ linux 6.8.0-90-generic (x64)                                                                                                      │
│ Node            │ 22.22.0                                                                                                                           │
│ Config          │ ~/.openclaw/openclaw.json                                                                                                         │
│ Dashboard       │ http://127.0.0.1:18789/                                                                                                           │
│ Tailscale       │ off                                                                                                                               │
│ Channel         │ stable (default)                                                                                                                  │
│ Update          │ pnpm · npm update 2026.2.19-2                                                                                                     │
│ Gateway         │ local · ws://127.0.0.1:18789 (local loopback) · reachable 42ms · auth token                                                       │
│ Security        │ Run: openclaw security audit --deep                                                                                               │
│ Gateway self    │ virt-5378 (127.0.0.1) app 2026.2.17 linux 6.8.0-90-generic                                                                    │
│ Gateway service │ systemd installed · enabled · running (pid 1581283)                                                                               │
│ Node service    │ systemd not installed                                                                                                             │
│ Agents          │ 7 total · 0 bootstrapping · 1 active · 1 sessions                                                                                 │
└─────────────────┴───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

Channels
┌──────────┬─────────┬────────┬───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Channel  │ Enabled │ State  │ Detail                                                                                                                │
├──────────┼─────────┼────────┼───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
└──────────┴─────────┴────────┴───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

Agents
┌────────────────────────┬───────────┬──────────┬──────────┬──────────────────────────────────────────────────────────────────────────────────────────┐
│ Agent                  │ Bootstrap │ Sessions │ Active   │ Store                                                                                    │
├────────────────────────┼───────────┼──────────┼──────────┼──────────────────────────────────────────────────────────────────────────────────────────┤
│ wazuh-triage           │ OK        │        0 │ unknown  │ ~/.openclaw/agents/wazuh-triage/sessions/sessions.json                                   │
│ wazuh-correlation      │ OK        │        1 │ 3m ago   │ ~/.openclaw/agents/wazuh-correlation/sessions/sessions.json                              │
│ wazuh-investigation    │ OK        │        0 │ unknown  │ ~/.openclaw/agents/wazuh-investigation/sessions/sessions.json                            │
│ wazuh-policy-guard     │ OK        │        0 │ unknown  │ ~/.openclaw/agents/wazuh-policy-guard/sessions/sessions.json                             │
│ wazuh-reporting        │ OK        │        0 │ unknown  │ ~/.openclaw/agents/wazuh-reporting/sessions/sessions.json                                │
│ wazuh-responder        │ OK        │        0 │ unknown  │ ~/.openclaw/agents/wazuh-responder/sessions/sessions.json                                │
│ wazuh-response-planner │ OK        │        0 │ unknown  │ ~/.openclaw/agents/wazuh-response-planner/sessions/sessions.json                         │
└────────────────────────┴───────────┴──────────┴──────────┴──────────────────────────────────────────────────────────────────────────────────────────┘

Diagnosis (read-only)

Gateway connection details:
  Gateway target: ws://127.0.0.1:18789
  Source: local loopback
  Config: /root/.openclaw/openclaw.json
  Bind: loopback

✓ Config: /root/.openclaw/openclaw.json
✓ Restart sentinel: none
! Port 18789
  Port 18789 is already in use.
  - pid 1581283 root: openclaw-gateway (127.0.0.1:18789)
  - Gateway already running locally. Stop it (openclaw gateway stop) or use a different port.
! Tailscale: off · unknown
  error: Error: spawn tailscale ENOENT
✓ Skills: 4 eligible · 0 missing · /root/.openclaw/wazuh-autopilot/agents/triage
✓ Channel issues (none)

Pasteable debug report. Auth tokens redacted.
Troubleshooting: https://docs.openclaw.ai/troubleshooting

i don't understand everything is running or not how to test and future proceed and how to validate everything is running. But i think i got MCP auth issue.

also on checking wazuh-mcp i get this

curl -X POST http://localhost:3000/mcp \
  -H "Authorization: Bearer dcf264d13b0107d835e077e029beb495dfeb692bbdb7331f8d5fa99781d55d76" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "tools/list", "id": 1}'

can you help above, and how to proceed further. Thanks

@nikopuf

[alokemajumder]

Hi @nikopuf,

Thanks for the detailed output — it helped us pinpoint the remaining issues. We've pushed fixes and here's what you need to do.

---

1. Config Validation Errors — Fixed

Both issues have been resolved in the latest commit:

• Model entries: Custom provider model definitions were missing required fields for the OpenClaw v2026.2.17 schema. Now corrected.
• Bindings: The peer objects in bindings were incomplete. Now corrected.

What to do:

git pull
cp openclaw/openclaw-airgapped.json ~/.openclaw/openclaw.json

If you're not using Slack, remove the channels and bindings sections entirely from your config before copying — agents work fully via REST API and heartbeats without Slack.

---

2. Gateway Mode

Your openclaw doctor output shows the gateway mode is unset. Fix with:

openclaw config set gateway.mode local
openclaw gateway restart

---

3. MCP Authentication

The /mcp endpoint uses session-based authentication — you need to initialize an MCP session first before calling tools/list or tools/call. Passing the API key directly as a Bearer token to /mcp without initializing won't work.

Next steps:

• Initialize an MCP session by sending an initialize request to /mcp with the required MCP protocol handshake (protocol version, client info)
• Use the session ID returned in the response header for subsequent tools/list and tools/call requests
• Your /health endpoint already confirms both Wazuh Manager and Indexer are healthy, so all 29 tools should be available once the session is established

---

4. Testing Without Slack

The entire two-tier approval workflow works via REST API — Slack is purely optional. You can:

1. Ingest alerts via POST /api/alerts on the runtime service (port 9090)
2. View cases via GET /api/cases
3. Create response plans via POST /api/plans
4. Approve plans (Tier 1) via POST /api/plans/{id}/approve
5. Execute plans (Tier 2) via POST /api/plans/{id}/execute

Check GET /health on port 9090 to confirm the runtime service is up.

---

Recommended Next Steps

1. git pull and copy the updated air-gapped config
2. Set gateway.mode to local
3. Restart the gateway and verify with openclaw status --all
4. Test the MCP connection using the session-based flow
5. Ingest a test alert via the REST API to validate the full pipeline

Let us know how it goes.