okf-memory
OKF Agent Memory is an independent open-source project that provides a domain-neutral, Git-native persistent memory layer for AI agents. It stores auditable Markdown/YAML knowledge in a repository and provides local BM25 retrieval, progressive disclosure, validation, and an embedded MCP server so coding agents can retain project context without an external vector database.
At a Glance
- Individual developers and software-engineering teams using AI coding agents
- Enterprises seeking auditable, vendor-neutral project memory
- Users of local/on-device LLMs such as Ollama and LM Studio
- Research, literature, coaching, and operations knowledge-base users
AI Tools by okf-memory
(1)
OKF Agent Memory
AI Agent Memory MCP Server
Discussions
No discussions yet
Be the first to start a discussion about okf-memory
Latest News
OKF Agent Memory v0.2.0 released with epistemic governance, code binding, scoped pre-edit discovery, dogfooding gates, and security hardening.
v0.1.5 released with adversarial security/DRY hardening, Google Jules integration, and boundary symlink containment.
v0.1.3 and v0.1.4 released for security hardening, Windows/cross-platform link handling, and OKF spec alignment.
v0.1.1 and v0.1.2 released with MCP JSON-RPC compliance, dynamic multi-bundle resolution, Homebrew automation, trust ordering, and stale validation.
Products & Services
An MIT-licensed, zero-external-dependency Go tool that initializes and bootstraps OKF v0.2 knowledge bundles; parses, validates, searches, shows, creates, updates, and relates Markdown concepts; and exposes native MCP tools over stdio.
The governance and code-binding release, with constraint/hold/context authority tiers, code_refs, scoped --for-path discovery, asset-drift safeguards, and hardened path and symlink boundaries.
Market Position
OKF positions itself between unstructured CLAUDE.md/AGENTS.md-style files and heavier vector-memory systems such as Mem0 and Letta (also discussed alongside Zep). Its differentiators are Git-native transparency, deterministic local BM25 retrieval, zero embedding/API cost, pure-Go low-latency execution, provenance/trust tiers, and code-aware governance; the trade-off versus vector systems is weaker semantic similarity retrieval.
Leadership
Founders
Stephan Knauer
Identified by contemporaneous coverage as the founder of okf-agent-memory; he uses the GitHub account sknr, authored the initial repository release, and is the copyright holder named in the MIT license. The project’s public materials do not provide a fuller employment or prior-company biography.
Founding Story
The project was started because AI coding agents lose architectural decisions, domain discoveries, and operational rules when context windows close or sessions reset. Its initial vision was an 'LLM Wiki' in pure Go: replace bloated CLAUDE.md/AGENTS.md files and opaque vector-memory services with structured, version-controlled Markdown that can be searched, reviewed, diffed, and rolled back.
Business Model
Revenue Model
The core CLI/library is distributed as open-source MIT-licensed software through source builds, Go install, precompiled GitHub releases, and a Homebrew tap. The website advertises an OKF Cloud beta waitlist, but no paid pricing or operating revenue is reported in the sources reviewed.
Target Markets
- Individual developers and software-engineering teams using AI coding agents
- Enterprises seeking auditable, vendor-neutral project memory
- Users of local/on-device LLMs such as Ollama and LM Studio
- Research, literature, coaching, and operations knowledge-base users
- Persistent architectural decisions and operational rules for autonomous software-engineering agents
- Claude Code, Cursor, Windsurf, Codex, Gemini CLI, and local-LLM coding workflows
- Research and literature-review knowledge bases
- Executive coaching and client-session knowledge bundles
- Operations and other domain-neutral project knowledge repositories