Files
Three primary files plus a changelog. Each is a stable URL — the _latest suffix always points at the current revision. When the registry changes, all three files are republished together.
Version history: changelog.md →
Not sure which to pick?
Three files, three jobs. Most users want the first.
- olm_system_prompt_latest.md — you're an educator or designer who wants to use Claude or ChatGPT to draft Core Mappings or Educator Briefs from program briefs.
- olm_context_bundle_latest.md — you're building an agent, tool, or evaluation that needs to understand OLM, not just produce content in it.
- canonical_registry_latest.yaml — you're writing code that validates OLM packets against the canonical vocabulary.
Quick usage
The fastest path is to copy the system prompt into your LLM and send a program brief. Three steps.
Open olm_system_prompt_latest.md and copy its contents into your LLM as a system message — or the first user message in tools without a system role.
Describe the learning experience you want to design — free text, no specific structure required. Duration, audience, what learners do and produce.
The model returns a Core Mapping in OLM's canonical vocabulary, ready to feed into a Playbook stage or to validate against the registry. If a concept in your brief has no canonical equivalent, the model notes the gap rather than inventing an ID.
For an end-to-end interactive flow with brief input and prompt assembly, the prompt builder on the main page wraps the system prompt with a task and your brief in one click.
Programmatic use
For agents, pipelines, or scripts, fetch the files directly. URLs are stable.
The repository at Meta-Humans/olm is the authoritative source. These files mirror canonical/canonical_registry.yaml and the AI assets on every release.
Why three files
OLM is a structural framework with a governed vocabulary. LLMs work well within it because the constraints are explicit — but only when the model can see the rules. The three primary files exist to give you three useful slices of the same source of truth.
The system prompt is the smallest workable unit. It contains the rules and registry an LLM needs to produce a Core Mapping or Playbook in canonical vocabulary, and nothing more. Use it for interactive work — pasting into a Claude or ChatGPT conversation.
The context bundle adds the constitution summary, the full layer contracts, and a worked example. It's larger, and it's the right choice when your tool or agent needs to reason about OLM rather than just generate within it. Evaluations, validators, and second-pass reviewers all benefit from the worked example.
The canonical registry YAML is the canonical list of valid IDs in a format your code can read. The repo's canonical/canonical_registry.yaml is the authoritative source; this file is its public mirror.