Skip to main content
Modern AI agents — whether you’re building your own or running one from the OpenClaw / MCP ecosystem — need different kinds of context at different moments in their execution. A planning step wants a system prompt and a couple of tool definitions. A retrieval step wants filtered document passages. A reasoning step might need structured data pulled from an integration. A reflection step wants all of the above plus recent conversation state. Wiring this up inside application code gets messy fast. Prompts live in one module, reference docs in a vector DB, connector data in another service, knowledge-base files loaded from disk at startup, and glue code stitching it all together. Every new agent or feature multiplies the surface area — and the bugs. Memic is a unified context layer for AI agents. One API key, one SDK, one interface for every kind of context your agent needs — unstructured documents, structured data from connectors, and managed prompts — all scoped to an isolated environment so tenants never cross-contaminate.

5-minute quickstart

Install the Python SDK and make your first search call.

Core concepts

Organizations, projects, environments, and folders.

API reference

The full REST API — 10 endpoints.

Python SDK

The memic package on PyPI.

​
What Memic unifies

  • Unstructured context — PDFs, Office docs, web pages, Markdown. Chunked, embedded, and semantically searchable.
  • Structured context — rows from connectors (Google Drive, Notion, S3, custom sources). Queryable alongside your documents.
  • Managed prompts — system prompts and templates stored in the dashboard, versioned, editable without a redeploy.
  • Multi-tenancy by design — each API key is locked to exactly one environment, so your agents never accidentally read a different customer’s context.

​
Why not just glue it together yourself?

Without a unified layer, a typical agent stack ends up touching:
  • prompts.py with hardcoded system prompts
  • Pinecone or Weaviate for vector search over docs
  • A separate indexer job that keeps vectors fresh as files change
  • Another system for structured data from Postgres / Notion / Salesforce
  • Glue code handling retries, pagination, auth, and errors
  • A tenancy layer you build yourself if customers need isolation
Every new feature extends one of these. Every new agent duplicates some of them. Every new customer means another isolation layer to maintain. Memic collapses all of it into a small API. You write client.search(query), client.prompts.get("planner-system"), or client.files.upload(...) — the underlying storage, retrieval, tenancy, and versioning is handled.

​
The shape of the API

curl https://api.memic.ai/api/v1/search \
  -H "X-API-Key: mk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"query": "what is our refund policy"}'
No org_id, no project_id, no environment_slug in the URL. Your API key already knows. Swap the key, swap the environment — your code doesn’t change.

​
Who Memic is for

  • Teams building AI agents that pull context from multiple sources — documents, databases, managed prompts — and don’t want to stitch three different systems together
  • Developers running agents from the OpenClaw / MCP ecosystem who need per-agent context isolation and versioned prompts
  • Platforms embedding RAG or agent features into their own products, where each customer’s context must stay isolated from every other customer’s

​
Ready to try it?

Start with the Python quickstart

Install, authenticate, upload a file, and search — in five minutes.