okf-agent-memory
An open-source memory management tool for AI agents that applies the package manager and lock file approach. It versions architectural decisions and domain knowledge as code dependencies and supports pull, vendor and lock workflows with static link verification.
Timeline (last 90 days)
okf-agent-memory is an open-source project that uses a 'package manager + lockfile' approach to manage AI Agent memory.
okf-agent-memory's workflow includes pulling dependencies via the CLI by executing 'okf pull <bundle>' or 'okf pull github.com/owner/repo@v1.0.0'.
After being pulled by OKF, the bundle is unpacked into the .okf/vendor/ directory and pinned with a SHA-256 hash in the okf.lock file.
OKF allows local project memory to directly reference concepts in vendor, such as @jwt/decisions/token-binding, and the linter statically verifies the validity of cross-bundle links.
OKF has no centralized account system, and anyone can publish a bundle by tagging a GitHub release with the 'okf-memory-bundle' topic.
The OKF project provides a web directory, registry.okf-memory.dev, as an entry point for browsing bundles.
The current toolchain version is okf-agent-memory v0.6.1.
The main changes in okf-agent-memory v0.6.1 include: native registry resolution, deterministic retrieval optimizations (fixing DF/TF scoring asymmetry, handling abbreviation suppression and prefix boundaries), and a persistence fix contributed by the community.
OKF claims to be a deterministic, local-first memory model, where the same input should yield the same retrieval results.
OKF is better suited to teams with clear architectural constraints and strong cross-repository reuse needs; if the knowledge itself is vague, exploratory, and without clear boundaries, forcing it into structure instead increases maintenance burden.
All Facts (10)
OKF is better suited to teams with clear architectural constraints and strong cross-repository reuse needs; if the knowledge itself is vague, exploratory, and without clear boundaries, forcing it into structure instead increases maintenance burden.
50%Unverifiedokf-agent-memory's workflow includes pulling dependencies via the CLI by executing 'okf pull <bundle>' or 'okf pull github.com/owner/repo@v1.0.0'.
50%UnverifiedAfter being pulled by OKF, the bundle is unpacked into the .okf/vendor/ directory and pinned with a SHA-256 hash in the okf.lock file.
50%UnverifiedOKF allows local project memory to directly reference concepts in vendor, such as @jwt/decisions/token-binding, and the linter statically verifies the validity of cross-bundle links.
50%UnverifiedOKF has no centralized account system, and anyone can publish a bundle by tagging a GitHub release with the 'okf-memory-bundle' topic.
50%UnverifiedThe OKF project provides a web directory, registry.okf-memory.dev, as an entry point for browsing bundles.
50%UnverifiedOKF claims to be a deterministic, local-first memory model, where the same input should yield the same retrieval results.
50%Unverifiedokf-agent-memory is an open-source project that uses a 'package manager + lockfile' approach to manage AI Agent memory.
50%UnverifiedThe current toolchain version is okf-agent-memory v0.6.1.
50%UnverifiedThe main changes in okf-agent-memory v0.6.1 include: native registry resolution, deterministic retrieval optimizations (fixing DF/TF scoring asymmetry, handling abbreviation suppression and prefix boundaries), and a persistence fix contributed by the community.
50%