Skip to content
Palimem Docsspec v1.7.0

Palimem and OKF

In mid-2026, Google published the Open Knowledge Format (OKF) v0.1 — an open specification that formalizes the “LLM wiki” pattern into a portable, vendor-neutral bundle of Markdown files with YAML frontmatter. No SDK required. Git-cloneable. Human-readable in any editor.

Palimem solves a different problem. OKF answers “what does this project know?” Palimem answers “what should the agent believe now, across sessions, with proof when it changed?”


OKF (Open Knowledge Format) is a directory of concept documents — each a Markdown file with YAML frontmatter declaring at minimum a type field. Optional reserved files like index.md and log.md provide structure and change history. Cross-links are ordinary Markdown links.

It is designed to be:

  • Authored by people or agents — curate domain knowledge, API references, playbooks
  • Exchanged without translation — any producer’s bundle can be consumed by any OKF-aware agent
  • Portable — tarball, git repo, or filesystem mount; no runtime server required

OKF does not specify supersession semantics, scoped runtime memory, search over governed state, or audit trails. It is a knowledge bundle format, not a memory server.


OKF Palimem
Question it answers What curated knowledge exists about this project or domain? What should the agent believe is current right now?
Format Markdown files + YAML frontmatter in a directory SQLite WAL + semantic units + MCP tools
Transport Files in git, tarball, or catalog MCP stdio server (local)
Updates Edit or regenerate markdown files memory_remember with supersession on conflict
Contradictions Multiple files can coexist; reader resolves Supersession resolves at write time per subject key
Scopes Not specified User, session, repository
Audit log.md changelog (optional) Immutable WAL, as_of recall, memory_audit_export
SDK required No MCP client (built into most coding agents)
Cloud required No No

Use OKF (or AGENTS.md, or a project wiki) when:

  • You want a curated, human-reviewable knowledge bundle about a codebase, API, or domain
  • Knowledge changes infrequently and is edited like documentation
  • You need something agents and humans can browse in git without a running server

Use Palimem when:

  • Facts change during agent sessions (stack choices, auth provider, deploy target)
  • You need one current value per subject — not the top-k similar chunks
  • You need scope isolation (user vs session vs repository)
  • You need to prove what was believed at a point in time

Use both when:

  • OKF (or AGENTS.md, llms.txt) carries stable project context the agent should read
  • Palimem carries session-to-session state that evolves — decisions made, facts learned, preferences stated

They compose. They do not replace each other.


A practical split:

Layer Tool Example content
Curated knowledge OKF bundle, AGENTS.md, or docs in git API shapes, architecture overview, coding conventions
Runtime memory Palimem MCP server “We switched auth to OAuth2 on 2026-07-15”, “User prefers tabs over spaces”, “Last deploy failed on lint step”

At session start, the agent reads your OKF bundle or repo instructions for stable context. During the session, Palimem captures governed facts via MCP tools (memory_remember, memory_search). When a fact changes, supersession keeps current truth on top — the OKF bundle does not need a rewrite for every session decision.

Palimem also publishes llms.txt and .well-known/ai-catalog.json for agent discovery — the same discovery pattern OKF bundles support, aimed at a different layer of the stack.


Format Role
OKF Curated knowledge bundle (markdown + frontmatter)
AGENTS.md Repo-level instructions for coding agents
llms.txt Agent discovery index for a site or project
Palimem Governed runtime memory server (MCP)

None of these are substitutes for the others. OKF and AGENTS.md describe what the project is. Palimem tracks what the agent believes now.