Any way to use this with Claude Code?

[seanGSISG]

Documentation is a bit over my head, do I really have to configure vLLM to use this with claude code?

[harshalmore31]

Hi @seanGSISG ! No, you don't need vLLM, but we don't directly support Claude Code or Anthropic models yet.

Current options:

Option 1 - Use Anthropic via LiteLLM:

from memori import Memori
from litellm import completion

memori = Memori(
    database_connect="sqlite:///memory.db",
    conscious_ingest=True,
    openai_api_key="your-openai-key"  # For memory processing
)
memori.enable()

# Use Claude via LiteLLM
response = completion(
    model="claude-3-sonnet-20240229",
    messages=[{"role": "user", "content": "Your prompt"}],
    api_key="your-anthropic-key"
)

Option 2 - OpenAI-compatible models:

memori = Memori(
    database_connect="sqlite:///memory.db",
    conscious_ingest=True,
    openai_api_key="your-openai-key"
)
memori.enable()

# Works with any OpenAI-compatible endpoint

What works now:
• OpenAI models (GPT-4, GPT-4o, etc.)
• Any OpenAI-compatible API
• Anthropic models through LiteLLM wrapper

Direct Anthropic/Claude Code support is on our roadmap, but not available yet. The LiteLLM approach should work for most Claude use cases though!

[colygon]

Totally fair question — and no, you don’t need to configure vLLM just to use Memori with Claude.

A key nuance: Memori is a Python SDK that integrates by registering against the LLM client object (Anthropic/OpenAI/etc.). Tools like Claude Code / Cursor / Cline are “agent shells” that often own the LLM call internally, so you can’t always wrap the internal client directly.

There are two practical integration patterns:

1. If you control the code that calls Claude (Anthropic): integrate Memori in-process by registering the Anthropic client, set attribution, and you’re done.
2. If Claude Code owns the LLM call: run a small local companion tool/service (HTTP or MCP) that exposes remember/recall endpoints backed by Memori, and configure the agent shell to call it.

I opened a docs PR with a concrete guide + examples for Claude Code / Cursor / Cline:
#286

If you share which tool you’re using (Claude Code vs Cursor vs Cline) and whether it supports MCP/tool calling in your setup, I can help you pick the shortest path.