Hindsight on Railway
Long-term memory server for AI agents, with REST, MCP and a web Control Plane, on Railway.
Hindsight is an open-source memory system for AI agents, built by Vectorize. Agents write memories to it and query them across sessions, using semantic, keyword, graph and time-aware search with reranking. It exposes a REST API and an MCP server, with Python, TypeScript and Go clients.
It is used by people building chatbots, support agents, coding assistants and multi-agent systems that should remember users and past decisions. With an LLM key it also extracts facts and entities from raw conversations and consolidates them into observations an agent can reflect on.
Hindsight is released under the MIT license. It stores everything in Postgres with the pgvector extension.
What Hindsight can do
- Semantic, keyword, graph and time-aware recall with reranking
- REST API and MCP endpoints per memory bank
- Background worker for retain, consolidation and imports
- Web Control Plane to browse banks, memories and entities
- Local embeddings and reranker, no external key required
- Optional LLM for fact extraction and reflect
What people use it for
- Long-term memory for chatbots and support agents
- Shared memory between agents through MCP
- Per-user or per-customer memory banks
- Inspecting and debugging what an agent knows
Videos
Running Hindsight on Railway
The Hindsight template deploys the API, a background worker, the Control Plane web UI and Postgres with pgvector in one project. The API and worker each hold the embedding and reranking models in memory, about 1 GB each at idle, which is the main cost driver; the template estimates roughly $20 a month idle and $45 to $50 with regular use, plus your LLM provider's charges. REST and MCP require an API key, and the Control Plane is protected by an access key.
Hindsight templates on Railway
Tracklayer templates are listed first. Other Hindsight templates follow, ordered by total deployments on Railway. How this works.
Deployment numbers as of 2026-11-20.
Questions about Hindsight
Do I need an LLM key?
No. Without one, memories are stored as chunks and recall works, but there is no fact extraction, consolidation or reflect. Add an LLM API key on the API service to enable them.
How do I authenticate?
Open the Control Plane with HINDSIGHT_CP_ACCESS_KEY. API and MCP requests need the tenant API key from the API service as a Bearer token.
How do I connect an MCP client?
Point it at https://{api-domain}/mcp/{bank}/ with the Authorization header set to your API key.
How do I upgrade?
Change the image tag on the API, worker and Control Plane together, back up Postgres first and redeploy; migrations run in the API's pre-deploy step.
Can I change the embedding model later?
Keep the embeddings provider fixed once memories exist. A different embedding size needs an empty database.
Guides
Similar apps
Tracklayer maintains the templates marked as such; Hindsight is a trademark of its owners and this site is not affiliated with them.