# RememberStack and remember.dev

> RememberStack is an open-source memory engine for AI agents. It keeps what each source said (claims), what is held true now (facts), when it was true, and the passage every answer came from. `remember` is its Python client, CLI and MCP server (`pip install remember`). remember.dev runs RememberStack for you, one isolated deployment per project.

Agents: resolve entities first (`resolve_entity`), then ask for facts (`facts_context`, with `time` for history), then claims and sources (`claims_and_sources_context`), then SQL queries over the query space. When the memory reports it knows nothing, say so; do not guess.

## Start

- [What is a memory system?](https://remember.dev/docs/start/what-is-a-memory-system): Why AI assistants forget, why searching your documents is not the same as remembering, and what a memory system does about it. No technical background needed.
- [Why RememberStack](https://remember.dev/docs): Memory for AI agents that keeps what each source said, what is true now, and when, with the path back to the source.
- [A five-minute tour](https://remember.dev/docs/start/how-it-works): What happens to a document from the moment you send it to the moment your agent asks about it.
- [remember.dev or self-hosted?](https://remember.dev/docs/start/choose): The same engine either way. What changes is who runs it.
- [Quickstart](https://remember.dev/docs/start/quickstart): Send a document to RememberStack and ask your first question.
- [Connect your coding agent](https://remember.dev/docs/start/connect-your-agent): Give Claude Code, Cursor, Codex, Antigravity or Claude Desktop access to your memory over MCP.

## Concepts

- [Documents, versions and sources](https://remember.dev/docs/concepts/documents-and-sources): How RememberStack identifies a file, keeps every version of it, and knows where it came from and when it was written.
- [Claims: what a source said](https://remember.dev/docs/concepts/claims): How RememberStack turns documents into standalone, source-grounded statements, what it leaves out, and why a claim never changes.
- [Facts: what is held true](https://remember.dev/docs/concepts/facts): How RememberStack combines claims from many sources into relations and observations, keeps them current, and counts their support.
- [Entities and identity](https://remember.dev/docs/concepts/entities): How RememberStack decides that two mentions are the same person, project or thing, and why a query never guesses between two.
- [Time](https://remember.dev/docs/concepts/time): How RememberStack records when something was true, when a source said it, and when the memory learned it, and how you ask about any of the three.
- [Evidence and provenance](https://remember.dev/docs/concepts/evidence): How every fact and claim in RememberStack points back to exact characters in a specific version of a source, and how to follow that path.
- [Contradictions, corroboration and supersession](https://remember.dev/docs/concepts/contradictions): How RememberStack shows disagreement between sources instead of hiding it, how it counts independent support, and how it tells a change from a conflict.
- [Updating a source: snapshot and living](https://remember.dev/docs/concepts/updating-sources): What a new version of a document means to RememberStack, and how a living source retracts facts when its content is removed.
- [The pipeline and readiness](https://remember.dev/docs/concepts/pipeline): What happens to a document between ingest and recall, stage by stage, how long it takes, and how to know when it can be queried.
- [What lives where](https://remember.dev/docs/concepts/architecture): The pieces of a RememberStack deployment, where each kind of data is kept, which copy is the authority, and why a stale search index can cost recall but never serve a withdrawn fact.
- [Retrieval: operations, search, graph and SQL](https://remember.dev/docs/concepts/retrieval): The ways to read from RememberStack, from four assured operations for everyday agent context to search, graph traversal and SQL queries, and how hybrid retrieval works underneath.
- [Reading a result](https://remember.dev/docs/concepts/reading-results): The envelope every RememberStack read returns, field by field, with an annotated example, and the ContextBundle/v2 wrapper.

## Guides

- [Supported file types](https://remember.dev/docs/guides/file-types): Which file types RememberStack reads on remember.dev and on a self-hosted deployment, and what to do with everything else.
- [Bring your existing data](https://remember.dev/docs/guides/bring-your-data): Move your existing files, vector-database contents and chat history into remember.dev or a self-hosted RememberStack, with their dates and sources intact.
- [Ingest files](https://remember.dev/docs/guides/ingest-files): Send files and raw bytes to RememberStack, name them so later edits land in the same document, and load a whole folder.
- [Ingest conversations and transcripts](https://remember.dev/docs/guides/ingest-conversations): Turn chat sessions and meeting transcripts into documents RememberStack can date, attribute and cite turn by turn.
- [Keep a source up to date](https://remember.dev/docs/guides/keep-sources-current): Re-send an edited file as a new version of the same document, and choose what happens to facts that the new version no longer supports.
- [Wait until a document is queryable](https://remember.dev/docs/guides/wait-for-readiness): Poll readiness for the versions you ingested, with a timeout that fits real processing times, and stop on failures instead of waiting forever.
- [Give an agent context](https://remember.dev/docs/guides/agent-context): Pick the assured operation that fits the question, size it to your token budget, and put its result into a prompt without losing what it tells you.
- [Ask about the past](https://remember.dev/docs/guides/ask-about-the-past): Ask what was true at a date, during a period, or ever, what sources said at the time, and what the memory believed on an earlier day.
- [Cite the source of an answer](https://remember.dev/docs/guides/cite-sources): Follow a fact to the claims behind it, the document they came from and the exact passage, and show why the memory holds it.
- [Handle unknowns and ambiguity](https://remember.dev/docs/guides/unknowns-and-ambiguity): Tell "nothing known" from "no such thing" from "could not look", handle names that match more than one entity, and brief your agent so it never fills a gap by guessing.
- [Explore memory with SQL](https://remember.dev/docs/guides/sql): Discover the query space, write your first SQL query over current facts, combine search with joins, and read the limits and errors.
- [Saved queries](https://remember.dev/docs/guides/saved-queries): List, inspect and run the 18 SQL queries every deployment ships with, and understand when a saved query runs and when it refuses.
- [Build a memory-backed agent](https://remember.dev/docs/guides/build-an-agent): A complete project assistant that loads a folder of meeting notes into RememberStack and answers questions with checked citations.

## remember.dev

- [What remember.dev runs for you](https://remember.dev/docs/cloud/overview): remember.dev runs a RememberStack deployment for each of your projects, so you send documents and ask questions without operating a database, a pipeline, model keys or TLS.
- [Organisations, projects and members](https://remember.dev/docs/cloud/organisations-and-projects): How remember.dev groups billing and people into an organisation, gives each project its own isolated deployment, and decides who can see and change what.
- [Tokens and sign-in](https://remember.dev/docs/cloud/tokens-and-sign-in): The credentials remember.dev issues, what each one can do, how long it lives, how to get it without copying secrets, and how to revoke it.
- [Hosted MCP](https://remember.dev/docs/cloud/hosted-mcp): Connect a coding agent to one remember.dev project over MCP with a browser sign-in instead of an API token, and what the hosted server offers.
- [Pricing and credits](https://remember.dev/docs/cloud/pricing): How remember.dev charges for an active project, for text it processes and for data it stores, how prepaid credits and holds work, and what is never charged.
- [Spend caps and auto top-up](https://remember.dev/docs/cloud/spend-controls): Set a monthly cap for the organisation and a sub-cap per project, understand exactly what stops when money or a cap runs out, and refill the balance automatically within a ceiling.
- [Limits](https://remember.dev/docs/cloud/limits): Every limit remember.dev enforces on ingest, storage, tokens, sign-in, projects and billing, with the error you get when you reach it.
- [Files and mounts](https://remember.dev/docs/cloud/files-and-mounts): Whether you can browse a remember.dev project's memory as files, and what to use instead.
- [What remember.dev serves](https://remember.dev/docs/cloud/compatibility): Which RememberStack version remember.dev runs, which deployment routes it serves, what differs from the version these docs describe, and how changes are announced.
- [Data handling and security](https://remember.dev/docs/cloud/data-and-security): Where remember.dev keeps your data, who and what can reach it, which outside services see it, what is recorded, and what remember.dev does not promise.
- [Closing a project, deleting your account](https://remember.dev/docs/cloud/leaving): How to close a remember.dev project and have its data deleted, how the 30-day cancel window works, how to delete your account, and what happens to unused credit.
- [Support](https://remember.dev/docs/cloud/support): How to reach remember.dev support, what to include so a problem can be traced, and what support can and cannot do today.

## Self-hosting

- [Requirements](https://remember.dev/docs/self-hosting/requirements): What you need to run RememberStack yourself — PostgreSQL 19, S3-compatible storage, Docker, a model key — and what is not measured yet.
- [Install with Docker Compose](https://remember.dev/docs/self-hosting/install): Start a self-hosted RememberStack with Docker Compose, check that it is healthy, send a first document, and stop, restart or reset it.
- [Configuration](https://remember.dev/docs/self-hosting/configuration): How a self-hosted RememberStack reads its settings, which ones you need to change, and how to set the ones compose.yaml does not pass through.
- [Models and providers](https://remember.dev/docs/self-hosting/models): Which model each pipeline step uses, how to change it, how RememberStack routes calls through OpenRouter, and what changing the embedding model does.
- [File formats and converters](https://remember.dev/docs/self-hosting/converters): Which file types a self-hosted RememberStack accepts, how to route more types to converters, and what happens to a file no converter accepts.
- [Authentication and scopes](https://remember.dev/docs/self-hosting/authentication): Protect a self-hosted RememberStack API with a shared secret or signed tokens, understand read, ingest and write scopes, and allow browser origins.
- [Scaling](https://remember.dev/docs/self-hosting/scaling): The bounds a self-hosted RememberStack runs within — worker claim rates, replicas per stage, connection pools for retrieval and graph queries, and SQL query limits.
- [Operating the pipeline](https://remember.dev/docs/self-hosting/operating): Inspect a self-hosted RememberStack pipeline, replay failed work, release parked conversions, set spend budgets, and repair the graph catalog.
- [Troubleshooting](https://remember.dev/docs/self-hosting/troubleshooting): Symptom, first check and fix for a self-hosted RememberStack that does not start, refuses requests, never finishes a document, or gives answers you did not expect.
- [Observability](https://remember.dev/docs/self-hosting/observability): What a self-hosted RememberStack logs, how to send its errors to Sentry or a compatible service, and how to export what it spent on models.
- [Upgrades and migrations](https://remember.dev/docs/self-hosting/upgrades): Move a self-hosted RememberStack to a new release, understand what setup migrates, keep the deployment id stable, and back up the data yourself.
- [Filesystem views](https://remember.dev/docs/self-hosting/filesystem-views): Publish a self-hosted RememberStack memory as a read-only directory tree that an agent can browse with ls, cat and grep.

## Reference

- [HTTP API conventions](https://remember.dev/docs/reference/http-api): Base URL, authentication, scopes, errors, limits and the other rules every RememberStack HTTP route follows.
- [Ingest, readiness and documents](https://remember.dev/docs/reference/http-api/ingest): The routes that add a document, report how far its processing has got, list what the deployment holds, and delete a document.
- [Assured operation routes](https://remember.dev/docs/reference/http-api/operations): List the four assured operations and run one over JSON arguments.
- [Entities and facts](https://remember.dev/docs/reference/http-api/entities-and-facts): Resolve a name to entities, look up relations and observations, hydrate a relation down to its sources, and read a relation's decision history.
- [Search and adjacent chunks](https://remember.dev/docs/reference/http-api/search): Search claims and source chunks by meaning or by keyword, and fetch the chunks around one you found.
- [Graph](https://remember.dev/docs/reference/http-api/graph): Walk the live graph of entities and relations, find paths between two entities, and trace citation chains between documents.
- [SQL queries](https://remember.dev/docs/reference/http-api/query): Run a read-only SQL statement over the memory_v1 query space, inspect its plan, discover the query space, and run saved queries.
- [Deployment info and health](https://remember.dev/docs/reference/http-api/deployment): Check which code and models a deployment is serving, probe its health, and read its cost export.
- [Assured operations](https://remember.dev/docs/reference/assured-operations): The four fixed retrievals RememberStack guarantees, with every parameter, limit, default and result contract.
- [Result types](https://remember.dev/docs/reference/result-types): Every response model of the RememberStack HTTP API, field by field, with every enum value.
- [Query space memory_v1](https://remember.dev/docs/reference/query-space): Every view, column, function, grammar rule, limit and shipped saved query of the memory_v1 query space that SQL queries run over.
- [Python SDK](https://remember.dev/docs/reference/python-sdk): Every public class, method, setting and error in the remember Python package.
- [CLI](https://remember.dev/docs/reference/cli): Every remember command, flag, default, environment variable and exit code.
- [MCP tools](https://remember.dev/docs/reference/mcp): The remember mcp server and the hosted remember.dev MCP server, every tool with its parameters, results and errors.
- [Configuration variables](https://remember.dev/docs/reference/configuration): Every environment variable read by the RememberStack engine and by the remember client and CLI, with its default, type and purpose.
- [Errors and status codes](https://remember.dev/docs/reference/errors): Every HTTP status, error code and status value RememberStack and remember.dev return, what each means, and what to do about it.
- [remember.dev API](https://remember.dev/docs/reference/cloud-api): The remember.dev account API for sign-in, organisations, projects, tokens, device login, billing, usage, settings, support, offboarding and hosted MCP sign-in, with authentication and error shapes.

## Project

- [Benchmarks and how we measure](https://remember.dev/docs/project/benchmarks): What LoCoMo and BEAM measure, how the RememberStack harness runs them, which numbers exist and for which protocol, and how to reproduce a run.
- [Releases and changelog](https://remember.dev/docs/project/changelog): What changed for users in each RememberStack release, newest first.
- [What is not built yet](https://remember.dev/docs/project/not-built-yet): Features RememberStack and remember.dev do not have today, each in one sentence.
- [Glossary](https://remember.dev/docs/project/glossary): Every term the RememberStack and remember.dev documentation uses, in one or two plain sentences, with a link to where it is explained.
- [Contributing, license and trademarks](https://remember.dev/docs/project/contributing): The license, the contributor agreement, the trademark policy, how to set up a development checkout and how the repository is laid out.
