The memory database for AI agents
MarkDB captures every chat your agents have, makes it searchable with hybrid vector and keyword retrieval, and auto-summarizes it -- all behind one OpenAI-compatible proxy. Point your coding agent at it and stop losing context.
OpenAI-compatible. No SDK lock-in. Bring your own model keys.
Drop-in for the tools your agents already use
Memory, search, and summaries -- without the plumbing
MarkDB sits between your agent and the model, capturing and organizing everything so you get recall and search for free.
Durable agent memory
Every prompt, tool call, and response is mirrored into a structured store -- organized into chats, sessions, and turns -- so nothing your agent did is ever lost.
Hybrid search
Retrieve past work with combined vector (pgvector) and full-text (Meilisearch) search. Semantic recall when you mean it, exact matches when you need them.
Automatic enrichment
A background worker summarizes turns, sessions, and whole chats so long histories collapse into compact, retrievable context instead of raw transcript noise.
OpenAI-compatible proxy
One endpoint speaks /v1/chat/completions, /v1/responses, /v1/messages, and /v1/embeddings, dispatching natively to Anthropic, OpenAI, and Gemini.
MCP server
Expose your agent's memory to any MCP-aware client for recall and search, without writing custom glue.
Bring your own keys
Per-tenant model credentials, scoped API keys, and members with roles. Consolidate spend on a single vendor or mix and match.
Three steps to durable agent memory
Mint an API key
Create a tenant in the dashboard and generate a scoped key (mk_live_...).
Point your agent at the proxy
Set your client's base URL to proxy.markdb.cloud. Cursor, Codex, and Claude Code each take two lines of config.
Get memory + search for free
Every exchange is captured, summarized, and indexed. Recall it from the dashboard, the API, or MCP.
Two lines to connect Claude Code
MarkDB speaks the Anthropic Messages API, the OpenAI Responses API, and Chat Completions. Pick your client and route it through the proxy.
# Point Claude Code at MarkDB, routed to Geminiexport ANTHROPIC_BASE_URL=https://proxy.markdb.cloudexport ANTHROPIC_AUTH_TOKEN=mk_live_xxx claude -p "summarize what we did yesterday" \ --model gemini-2.5-flashGive your agents a memory that lasts
Start capturing and searching your agent history in minutes.