# OMEM

> Open-source (MIT) memory for AI agents built on belief revision. OMEM keeps
> both sides of a contradiction, tracks what an agent believed and when, answers
> "why" with a provenance chain, and puts a human approval step before a risky
> action runs. Self-hosted, one `pip install omem-infrastructure`, no
> dependencies, dashboard included. It phones home to nobody (CI-asserted).

OMEM is for two audiences: builders who want agent memory that never silently
overwrites a contradiction, and teams shipping agents to clients who must answer
"why did the agent do that" with an audit trail and gate risky actions behind a
named human approver. The engine is frozen and replays byte-identically, so an
upgrade can never rewrite recorded history. Its wider objective: teach AI what
people are like while holding a fact about no one, via anonymous behavioural
priors (counts, never people) that consenting installs pool into a training
corpus.

## Docs

- [Quickstart](https://infrastructure.omem-cloud.com/docs/quickstart/): install and first belief in about a minute
- [Documentation](https://infrastructure.omem-cloud.com/docs/): the memory model, verbs, and concepts
- [SDK reference](https://infrastructure.omem-cloud.com/docs/sdk/): the nine verbs of the Python SDK

## Product

- [Accountability](https://infrastructure.omem-cloud.com/accountability/): prove why the agent acted, approve before it does
- [Objective](https://infrastructure.omem-cloud.com/objectives/): teach AI what people are like, holding a fact about no one
- [Security](https://infrastructure.omem-cloud.com/security/): what is implemented today and what is not built yet
- [Pricing](https://infrastructure.omem-cloud.com/pricing/): free and MIT while in beta

## Specification

- [The Testimony Record Specification v0.1](https://infrastructure.omem-cloud.com/spec/testimony-record/): the open format for what an AI agent believed, what it was allowed to do, and who allowed it; four conformance levels, free to implement
- [Testimony Record implementations](https://infrastructure.omem-cloud.com/spec/testimony-record/implementations): which systems emit conforming records, the level each reaches, and how to get listed; listings follow a record that passed the validator, never self-report
- [JSON Schema](https://infrastructure.omem-cloud.com/testimony-record-v0.1.schema.json)

## Comparisons

- [OMEM vs Mem0 vs Zep, side by side](https://infrastructure.omem-cloud.com/compare/omem-vs-mem0-vs-zep/): the three-way table, with what each optimizes and where each honestly wins
- [OMEM vs Mem0](https://infrastructure.omem-cloud.com/compare/mem0/)
- [OMEM vs Zep](https://infrastructure.omem-cloud.com/compare/zep/)
- [OMEM vs Letta](https://infrastructure.omem-cloud.com/compare/letta/)

## Guides

- [EU AI Act Article 12 for AI agents](https://infrastructure.omem-cloud.com/guides/eu-ai-act-article-12-ai-agents/): the high-risk logging duty as engineering, with working code
- [An audit trail for AI agents](https://infrastructure.omem-cloud.com/guides/ai-agent-audit-trail/): what a defensible trail must contain, and why request logs do not qualify
- [Human-in-the-loop approvals for AI agents](https://infrastructure.omem-cloud.com/guides/human-in-the-loop-ai-agents/): gating risky actions behind a named approver, with refusals on the record
- [Should an agent's memory decide what is true?](https://infrastructure.omem-cloud.com/guides/should-agent-memory-decide-truth/): the essay behind the design
- [Agent memory in Python](https://infrastructure.omem-cloud.com/guides/agent-memory-python/)
- [LangGraph long-term memory](https://infrastructure.omem-cloud.com/guides/langgraph-long-term-memory/)
- [An MCP memory server](https://infrastructure.omem-cloud.com/guides/mcp-memory-server/)

## Full content

- [Complete docs and guides in one file](https://infrastructure.omem-cloud.com/llms-full.txt): the full text of every doc, guide, and comparison page, for AI assistants that want deep context without crawling

## Source

- [GitHub repository](https://github.com/troybrandonc-bit/Omem): MIT source, claims ledger, Witness benchmark
- [PyPI package](https://pypi.org/project/omem-infrastructure/): pip install omem-infrastructure
