Recall
Memory and resumable state for AI agents, with every correction kept as history
- Pricing
- Open source
- Platforms
- CLI, API, self-hosted
- Version
- 0.25.0
- Listing updated
- today
Similar MCP servers
About Recall
Recall is an open-source MCP server for knowledge and memory. It is available on CLI, API, self-hosted and Web.
The README presents Recall as a way to give agents durable memory that outlives a single conversation or process. Facts a user states are stored with every correction preserved, so an agent can report what it believes now, what it believed earlier, and the source of each belief. Separately, a state layer records the agent's goal, plan, progress, notes and in-flight tool calls, so a crashed or relocated run can pick up from its own records.
Recall ships as a Go library, a recall binary that exposes the same operations over MCP and HTTP, and a Python client. The README says data is kept in a database the user runs, including polign_db, PostgreSQL, MongoDB or Qdrant, with a local directory as the default. Its source is published under the Apache 2.0 licence.
The project suits developers building agents that must remember user preferences across sessions, or that need to recover after a failure without repeating side-effecting actions.
What it does
- Correction history
- Each change to a remembered fact is recorded, and queries can return what was believed at a given date.
- Source tracing
- The explain command shows which message a fact came from, the rule that applied, and its history.
- Crash-safe resume
- Working state, recent turns and unfinished tool calls are restored so a new process continues the run.
- Typed predicate registry
- Predicates define value types, single or multiple values, and which predicates they contradict.
- Pluggable storage
- Memory can be kept in a local directory, polign_db, PostgreSQL with pgvector, MongoDB or Qdrant.
- Multi-agent HTTP server
- A serve mode hosts memory and state for many agents over HTTP with per-namespace isolation.
Good for
- Keeping a coding agent aware of a user's stated preferences across sessions
- Answering what an assistant believed at an earlier date after the user changes a fact
- Resuming a long-running agent task after a process crash without repeating tool calls
- Sharing a common memory store across several agents on one machine or in a cluster
Pricing
Open source
Apache 2.0 licensed, self-hosted on a database you run.
Recall FAQ
How do I install Recall?
The README lists pip install polign-recall, a Homebrew formula, and go install for the binary alone.
Where is memory stored by default?
With no configuration, memory is kept in a local database in ~/.local/share/recall, shared by agents on the machine.
What license does Recall use?
The README states Apache 2.0.
Does Recall need an API key or model to start?
The README says no model and no API key are needed to start; a model is only needed for plain-text extraction.
Is Recall free?
Yes. Recall is open source, so you can use and self-host it for free.
What are the best alternatives to Recall?
The closest alternatives to Recall are SuperLocalMemory, Total Agent Memory, GrayMatter, Heartwood Memory — governed memory for AI agents and Oort.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.