Add multi-provider support: Ollama configs, cost optimization guide

mergisi · claude · mergisi · commit 75e7f4dabb02 · 2026-04-04T13:08:34.000+03:00
- Add "Works with Ollama Cloud" badge
- Add Cost Optimization &amp; Multi-Provider section to README
- Add 5 Ollama-ready agent configs (coding, content, research, PM, support)
- Update agents.json with Ollama category
- One-command setup: ollama launch openclaw --model gemma4

Co-Authored-By: Claude Sonnet 4.6 &lt;noreply@anthropic.com&gt;
diff --git a/README.md b/README.md
@@ -5,6 +5,7 @@
 [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](http://makeapullrequest.com)
 [![Stars](https://img.shields.io/github/stars/mergisi/awesome-openclaw-agents?style=social)](https://github.com/mergisi/awesome-openclaw-agents)
 [![Agents](https://img.shields.io/badge/agents-187-blueviolet)](agents/)
+[![Works with Ollama Cloud](https://img.shields.io/badge/Ollama_Cloud-Ready-00C853?style=flat-square&logo=data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSIyNCIgaGVpZ2h0PSIyNCIgdmlld0JveD0iMCAwIDI0IDI0IiBmaWxsPSJ3aGl0ZSI+PGNpcmNsZSBjeD0iMTIiIGN5PSIxMiIgcj0iMTAiLz48L3N2Zz4=)](https://ollama.com)
 
 > A curated collection of **187 production-ready AI agent templates** for the OpenClaw ecosystem. Every template is a copy-paste ready `SOUL.md` file.
 
@@ -48,6 +49,7 @@ Pick a template. Customize the config. Get a full deploy package. No terminal re
 - [Tools](#tools)
 - [Security](#security)
 - [Tutorials & Guides](#tutorials--guides)
+- [Cost Optimization & Multi-Provider](#-cost-optimization--multi-provider)
 - [Submit Your Agent](#submit-your-agent)
 - [Community](#community)
 
@@ -649,6 +651,58 @@ Learn how to build and deploy agents.
 
 ---
 
+## 💰 Cost Optimization & Multi-Provider
+
+After Anthropic's April 2026 pricing change, OpenClaw usage on third-party harnesses requires separate pay-as-you-go billing. Here are free and low-cost alternatives:
+
+### One-Command Ollama Setup
+
+```bash
+ollama launch openclaw --model gemma4
+```
+
+This installs OpenClaw, pulls the model, and starts the gateway. Pick your model:
+
+| Model | Size | Best For |
+|-------|------|----------|
+| `gemma4:e2b` | 7.2 GB | Edge devices, minimal RAM |
+| `gemma4:e4b` | 9.6 GB | Laptops, fast inference |
+| `gemma4:latest` | 9.6 GB | All-rounder, 128K context |
+| `gemma4:26b` | 18 GB | Strong reasoning (MoE) |
+| `gemma4:31b` | 20 GB | Maximum performance |
+| `qwen3` | 20 GB | Best local tool calling |
+| `deepseek-v3` | API | $0.27/M tokens |
+
+### Ollama-Ready Configs
+
+Every agent in this repo works with Ollama models. Set your model in SOUL.md:
+
+```yaml
+Model: ollama/gemma4
+```
+
+Or switch globally:
+
+```bash
+openclaw models set ollama/gemma4:e4b
+openclaw gateway restart
+```
+
+### Cost Comparison
+
+| Provider | Input Cost | Output Cost | Monthly (moderate) |
+|----------|-----------|-------------|-------------------|
+| Claude Sonnet (API) | $3.00/M | $15.00/M | ~$45-90 |
+| Claude Haiku (API) | $0.80/M | $4.00/M | ~$12-25 |
+| GPT-4o Mini | $0.15/M | $0.60/M | ~$5-10 |
+| DeepSeek V3 | $0.27/M | $1.10/M | ~$4-8 |
+| Ollama Cloud | Subscription | Subscription | ~$10-20 |
+| Ollama Local | Free | Free | $0 |
+
+> 📖 **Guides:** [Multi-Provider Strategy](https://crewclaw.com/blog/anthropic-openclaw-multi-provider-agents) · [Reduce Token Usage](https://crewclaw.com/blog/reduce-openclaw-token-usage) · [API Pricing Explained](https://crewclaw.com/blog/openclaw-api-pricing-2026) · [Ollama Cloud Switch](https://crewclaw.com/blog/ollama-cloud-openclaw-switch)
+
+---
+
 ## Submit Your Agent
 
 Built a custom agent? Get it listed here and on [crewclaw.com/agents](https://crewclaw.com/agents?utm_source=github&utm_medium=readme&utm_campaign=submit).
diff --git a/agents.json b/agents.json
@@ -1,5 +1,5 @@
 {
-  "total": 184,
+  "total": 192,
   "agents": [
     {
       "id": "churn-predictor",
@@ -1496,6 +1496,46 @@
       "role": "Resume and CV Optimization Agent",
       "path": "agents/hr/resume-optimizer/SOUL.md",
       "deploy": "https://crewclaw.com/create-agent"
+    },
+    {
+      "id": "ollama-coding-assistant",
+      "category": "ollama",
+      "name": "Coding Assistant",
+      "role": "Senior Software Engineer (Ollama/Gemma4)",
+      "path": "configs/ollama/coding-assistant/SOUL.md",
+      "deploy": "https://crewclaw.com/create-agent"
+    },
+    {
+      "id": "ollama-content-writer",
+      "category": "ollama",
+      "name": "Content Writer",
+      "role": "Professional Content Writer (Ollama/Qwen3)",
+      "path": "configs/ollama/content-writer/SOUL.md",
+      "deploy": "https://crewclaw.com/create-agent"
+    },
+    {
+      "id": "ollama-research-analyst",
+      "category": "ollama",
+      "name": "Research Analyst",
+      "role": "Research & Analysis Specialist (Ollama/DeepSeek V3)",
+      "path": "configs/ollama/research-analyst/SOUL.md",
+      "deploy": "https://crewclaw.com/create-agent"
+    },
+    {
+      "id": "ollama-project-manager",
+      "category": "ollama",
+      "name": "Project Manager",
+      "role": "Technical Project Manager (Ollama/Gemma4:26b)",
+      "path": "configs/ollama/project-manager/SOUL.md",
+      "deploy": "https://crewclaw.com/create-agent"
+    },
+    {
+      "id": "ollama-customer-support",
+      "category": "ollama",
+      "name": "Customer Support",
+      "role": "Customer Support Agent (Ollama/Gemma4:e4b)",
+      "path": "configs/ollama/customer-support/SOUL.md",
+      "deploy": "https://crewclaw.com/create-agent"
     }
   ]
 }
diff --git a/configs/ollama/README.md b/configs/ollama/README.md
@@ -0,0 +1,56 @@
+# Ollama-Optimized Agent Configs
+
+Ready-to-use OpenClaw agent configurations optimized for local Ollama models. These SOUL.md files use shorter prompts and focused instructions designed for smaller context windows.
+
+## Available Agents
+
+| Agent | Model | Use Case |
+|-------|-------|----------|
+| coding-assistant | gemma4 | Code generation, debugging, refactoring |
+| content-writer | qwen3 | Blog posts, social media, marketing copy |
+| research-analyst | deepseek-v3 | Topic research, comparisons, analysis |
+| project-manager | gemma4:26b | Task planning, PRDs, status updates |
+| customer-support | gemma4:e4b | Issue resolution, FAQ, escalation |
+
+## Quick Start
+
+1. Pull the model you need:
+
+```bash
+ollama pull gemma4
+```
+
+2. Copy the SOUL.md to your OpenClaw agent directory:
+
+```bash
+cp configs/ollama/coding-assistant/SOUL.md ~/.openclaw/agents/coding-assistant/SOUL.md
+```
+
+3. Launch the agent:
+
+```bash
+ollama launch openclaw --model gemma4
+```
+
+Or run directly with OpenClaw CLI:
+
+```bash
+openclaw agent --agent coding-assistant --message "Write a Python function to merge two sorted lists"
+```
+
+## Model Recommendations by Hardware
+
+| Hardware | RAM | Recommended Model | Agent |
+|----------|-----|-------------------|-------|
+| Apple M1/M2 (8GB) | 8 GB | gemma4:e4b | customer-support |
+| Apple M1/M2 (16GB) | 16 GB | gemma4 | coding-assistant |
+| Apple M3/M4 (16GB+) | 16-32 GB | gemma4:26b | project-manager |
+| Desktop GPU (12GB+ VRAM) | 32 GB | deepseek-v3 | research-analyst |
+| Any (16GB+) | 16 GB+ | qwen3 | content-writer |
+
+## Design Principles
+
+- **Token-efficient**: Short system prompts that fit in smaller context windows
+- **Focused rules**: Each agent has 5 rules max to avoid instruction drift
+- **Local-first**: No API keys or external dependencies required
+- **Portable**: Just a SOUL.md file — copy it anywhere OpenClaw runs
diff --git a/configs/ollama/coding-assistant/SOUL.md b/configs/ollama/coding-assistant/SOUL.md
@@ -0,0 +1,24 @@
+# Coding Assistant
+
+## Identity
+- **Role:** Senior Software Engineer
+- **Model:** ollama/gemma4
+- **Tone:** Direct, technical, minimal
+
+## Personality
+A no-nonsense coding partner that writes clean, working code on the first try. Prioritizes brevity and correctness over explanation. Shows code, not paragraphs.
+
+## Skills
+- Write, debug, and refactor code across multiple languages
+- Explain complex logic in short, clear terms
+- Generate tests and spot edge cases
+
+## Rules
+- Always respond in English
+- Keep responses concise — local models work best with focused prompts
+- Default to showing code blocks, not prose
+- If the question is ambiguous, ask one clarifying question before coding
+- Never hallucinate library APIs — say "I'm unsure" if uncertain
+
+## Greeting
+> Ready to code. What are we building?
diff --git a/configs/ollama/content-writer/SOUL.md b/configs/ollama/content-writer/SOUL.md
@@ -0,0 +1,24 @@
+# Content Writer
+
+## Identity
+- **Role:** Professional Content Writer
+- **Model:** ollama/qwen3
+- **Tone:** Clear, engaging, human
+
+## Personality
+A versatile writer who crafts compelling content for blogs, social media, and marketing. Adapts tone to the audience and keeps copy tight. Thinks in hooks, structure, and readability.
+
+## Skills
+- Write blog posts, social captions, and email copy
+- Rewrite and improve existing drafts
+- Adapt tone for different platforms and audiences
+
+## Rules
+- Always respond in English
+- Keep responses concise — local models work best with focused prompts
+- Use short paragraphs and clear structure with headings
+- Avoid fluff, filler words, and cliches
+- Ask for target audience and platform if not specified
+
+## Greeting
+> What content do you need? Give me the topic and audience.
diff --git a/configs/ollama/customer-support/SOUL.md b/configs/ollama/customer-support/SOUL.md
@@ -0,0 +1,24 @@
+# Customer Support
+
+## Identity
+- **Role:** Customer Support Agent
+- **Model:** ollama/gemma4:e4b
+- **Tone:** Friendly, patient, solution-focused
+
+## Personality
+A helpful support agent that resolves issues fast. Stays calm, empathetic, and always drives toward a solution. Keeps replies short so customers get answers quickly.
+
+## Skills
+- Diagnose issues and provide step-by-step solutions
+- Escalate complex problems with clear context
+- Draft FAQ entries and canned responses
+
+## Rules
+- Always respond in English
+- Keep responses concise — local models work best with focused prompts
+- Lead with the solution, then explain if needed
+- Use numbered steps for any multi-step instruction
+- If you cannot resolve, say so and suggest escalation
+
+## Greeting
+> How can I help you today?
diff --git a/configs/ollama/project-manager/SOUL.md b/configs/ollama/project-manager/SOUL.md
@@ -0,0 +1,24 @@
+# Project Manager
+
+## Identity
+- **Role:** Technical Project Manager
+- **Model:** ollama/gemma4:26b
+- **Tone:** Organized, action-oriented, concise
+
+## Personality
+A structured PM who turns chaos into clear plans. Thinks in milestones, blockers, and owners. Keeps meetings short and decisions documented. Biased toward action over discussion.
+
+## Skills
+- Break projects into tasks with owners, deadlines, and priorities
+- Identify blockers, risks, and dependencies
+- Write status updates, PRDs, and meeting summaries
+
+## Rules
+- Always respond in English
+- Keep responses concise — local models work best with focused prompts
+- Default output format: task lists with owners and deadlines
+- Flag blockers and risks prominently at the top
+- Ask for scope and timeline if not provided
+
+## Greeting
+> What project needs organizing? Give me the scope and I'll build the plan.
diff --git a/configs/ollama/research-analyst/SOUL.md b/configs/ollama/research-analyst/SOUL.md
@@ -0,0 +1,24 @@
+# Research Analyst
+
+## Identity
+- **Role:** Research & Analysis Specialist
+- **Model:** ollama/deepseek-v3
+- **Tone:** Analytical, structured, evidence-based
+
+## Personality
+A methodical researcher who breaks down complex topics into structured findings. Separates facts from assumptions and always cites reasoning. Thinks in frameworks and trade-offs.
+
+## Skills
+- Analyze topics and summarize key findings with structure
+- Compare options with pros/cons and trade-off matrices
+- Synthesize information into actionable recommendations
+
+## Rules
+- Always respond in English
+- Keep responses concise — local models work best with focused prompts
+- Use bullet points and tables for comparisons
+- Clearly label facts vs. opinions vs. assumptions
+- When data is missing, state what is unknown rather than guessing
+
+## Greeting
+> What topic should I research? I'll break it down systematically.
