contextos

Your AI tools have different brains. ContextOS gives them one.

Claude remembers Claude. ChatGPT remembers ChatGPT. Your custom agent remembers nothing. ContextOS is the shared memory layer that runs on your machine — any LLM reads from it, any LLM writes to it.

Install

pip install "contextos[cli]"

One-time setup

contextos init

This starts the server, creates an API key, and saves config to ~/.contextos/config.json. Takes 30 seconds.

Add 2 lines to your existing app

import contextos
import anthropic

contextos.init()             # reads config saved by `contextos init`
contextos.set_user("alice")  # set before each LLM call

# Your existing code — UNCHANGED
response = anthropic.Anthropic().messages.create(
    model="claude-opus-4-6",
    system="You are helpful.",
    messages=[{"role": "user", "content": message}]
)
# ContextOS automatically:
#   → injects Alice's memory into the system prompt
#   → captures the conversation in the background

Works with OpenAI too — zero changes to your openai.OpenAI().chat.completions.create(...) calls.

Manual API (full control)

# In your Claude app
claude_client = ContextOS(api_key="sk-...")
claude_client.write(user_id="alice", conversation="...", source_client="claude-app")

# In your GPT app — reads memory written by the Claude app
gpt_client = ContextOS(api_key="sk-...")
memory = gpt_client.query(user_id="alice", q="what does alice prefer?")
# → Alice never re-introduced herself. Your GPT app already knows her.

Same user_id. Same server. One brain.

Create an API key

contextos keys create --app-name myapp \
  --database-url postgresql://contextos:contextos@localhost:5433/contextos

API

client.write(user_id, conversation, source_client=None)   # → session_id
client.query(user_id, q, top_k=10, scope="global")        # → MemoryResponse
client.delete(fragment_id)

# Async versions
await client.awrite(...)
await client.aquery(...)
await client.adelete(...)

MemoryResponse.prompt_block is a pre-formatted string you paste directly into your system prompt. No processing needed.

CLI

contextos start          # start server
contextos stop           # stop server
contextos logs -f        # follow logs
contextos health         # check status
contextos keys create    # create API key
contextos keys list      # list apps

Links

• GitHub
• Docs
• MIT License