Knowledge transfer
Preserve craft when people leave
A senior cobbler retires; a new apprentice opens the same Procedure Pack and learns the resoling sequence on their phone. The craft survives the generation.
An open standard for operational knowledge
The world's procedures finally writable.
Most of how work actually gets done lives in people's heads. OpenPSL is the open standard — and the platform — that turns that operational knowledge into something writable, shareable, and machine-readable. For humans, for teams, for AI agents.
Open source Free to use Language-agnostic Machine-readable
The problem
How work actually happens is mostly undocumented — and what little documentation exists is a mess.
Knowledge lives in heads
When a senior worker retires, leaves, or gets sick, their procedures often leave with them. Trades, crafts and institutional know-how disappear quietly.
Every team reinvents the wheel
Each organization writes its own onboarding, its own incident response, its own intake forms. The same procedures get redrafted thousands of times, badly.
No shared format
Wikis, PDFs, Word docs, spreadsheets, sticky notes. No two organizations document procedures the same way. Nothing interoperates.
Audit and compliance suffer
When a regulator asks for 'the procedure', you cobble together evidence. Versions diverge. Reality drifts from policy.
AI agents can't read it
LLMs can summarize prose, but they can't reliably automate work without structured, machine-readable procedures to follow.
Apprentices learn the hard way
New workers shadow seniors for weeks because there's no good way to read the job. Onboarding costs balloon, quality varies.
The standard
One schema. Built for humans. Readable by machines.
OpenPSL — the Open Procedure Specification Language — defines a tiny universal schema for any procedure, anywhere in the world. Twelve fields. Nothing exotic. Extensible without breaking.
The shape of every procedure
title
what
industry
sector
domain
sub-area
objective
why
roles
who
inputs
needed before
steps
ordered actions
decision_rules
if/then · matrix · formula · fallback
outputs
produced after
exceptions
edge cases
risks
what could go wrong
controls
how to mitigate
Nine sections are required (title, industry, domain, objective, roles, inputs, steps, decisions, outputs). Exceptions, risks and controls are optional — add them when they fit your procedure.
What the editor produces under the hood
{
"title": "Daily harvest readiness check",
"objective": "Decide whether today justifies a pick.",
"industry": "Agriculture",
"domain": "Coffee harvesting",
"roles": [
{ "name": "Farm owner",
"description": "Makes the final call." }
],
"inputs": [
{ "name": "Weather forecast",
"roles": ["Farm owner"] },
{ "name": "Sample baskets",
"roles": ["Farm owner"] }
],
"steps": [
{ "name": "Walk three random rows",
"roles": ["Farm owner"],
"estimatedTime": "20 min" },
{ "name": "Check weather forecast",
"roles": ["Farm owner"] }
],
"decision_rules": [
{ "type": "if_then",
"condition": "Red cherries >= 70% AND no rain in 8h",
"ifTrue": "Start picking by 7:00",
"ifFalse": "Delay 24h and recheck" }
],
"outputs": [
{ "name": "Pick / no-pick decision" }
],
"risks": [
{ "name": "Picking under-ripe cherries",
"likelihood": "medium", "impact": "high",
"roles": ["Farm owner"] }
],
"controls": [
{ "name": "Three-row sampling rule",
"mitigates": ["riskId-cherry-ripeness"],
"roles": ["Farm owner"] }
]
}Minimal
Fewer than a dozen fields cover any procedure on the planet.
Open
JSON Schema, MIT-licensed implementation, no vendor lock-in.
Universal
Works in any language, any industry, any device. Mobile-first by design.
The same pack, three lives.
A single Procedure Pack becomes a document, a checklist, an API payload, and training data — without anyone copying and pasting.
1
Humans
Operators read it as a clean web page or printable PDF. Apprentices follow it step by step on mobile.
2
Systems
ERP, BPM, ticketing or QA tools ingest the JSON and trigger workflows or audit checks.
3
AI agents
LLM agents read the pack as ground truth and execute, evaluate or coach against it.
Why now
The missing layer for AI in the workplace.
Large language models are remarkable. They can summarize, draft and chat. What they can't do is reliably act on operational work — because the operational work isn't written down in a form they can follow.
Every team building an AI agent today is rebuilding the same thing: a janky prompt that tries to encode a procedure. OpenPSL replaces those prompts with structured, versionable, auditable Procedure Packs. One source of truth that humans, systems and agents share.
Without a standard
Companies hand their AI agents free-form documents, screenshots, and 8,000-token prompts. The agent guesses, hallucinates, and drifts. Each new use case means rewriting the prompt from scratch.
With OpenPSL
Agents read a tiny, structured JSON. They follow the same steps a human would. They report back against the same decision rules. Procedures evolve in one place; every agent and every operator stays in sync.
Agent-ready
Feed a pack to any LLM and it has everything it needs to act.
Auditable
Every action maps to a step, role, decision. No more black box.
Reusable
Train one model, deploy across teams and geographies.
Versioned
Procedures evolve safely. Agents follow the same version your team does.
Real use cases
Where structured procedures actually pay off.
Not theory. Concrete scenarios where a single pack outperforms a stack of PDFs, a wiki, or a tribal habit.
Onboarding
Turn 6 weeks of shadowing into 2
A clinic captures every front-desk procedure as packs. New hires read, run and refine them in their first shifts instead of memorizing rituals.
AI automation
Ground your agents in real procedures
A bank feeds its KYC, AML and reconciliation packs to an LLM agent. The agent triages alerts following the exact same rules the human team uses — versioned and auditable.
Compliance & audit
One structured source for every auditor
An auditor asks 'show me your incident response procedure'. You hand over a versioned JSON pack plus the human-readable PDF. Same content, same date, no Word document drift.
Cross-team alignment
Same procedure, local forks
A logistics company runs the same outbound-dispatch pack across 12 hubs. Each hub forks for its quirks. The mother pack stays canonical; hubs sync improvements upstream.
Process improvement
See exactly what changed
A factory line updates its changeover procedure to cut down setup time. Version diff shows the before/after; quality team understands what's new without a meeting.
Vocational education
Real curricula, not theory
A trade school builds courses around community packs in electrical, plumbing, automotive. Students follow real-world procedures from week one, not generic theory.
Public services
Civic procedures readable by all
A government agency publishes citizen-service procedures so AI assistants can guide citizens through them — and so the procedures themselves become legible to the public.
Industry consortia
Shared best practices as a public good
A trade association publishes a pack library for safety, quality and onboarding. Members fork, adapt and contribute back. The whole industry levels up.
Who benefits
Value at every layer — from one person to the planet.
For individuals
Capture your craft. Build a personal procedure library that follows you across jobs and decades. Document what you know before you forget. Share with the world if you want to.
For teams
Stop rewriting the same onboarding doc, the same incident response, the same intake form. Align on one structured source. Diff versions to see what changed. Reuse packs across projects.
For organizations
Build a private library of how your operations actually run. Pipe packs into your BPM, ERP, training and audit systems. Get a defensible source of truth for compliance and continuity.
For AI builders
Train and evaluate agents against structured procedures, not raw prompts. Get reliable, auditable automation. Build agents that follow the same procedure version your team is reading.
For educators & apprentices
Teach real procedures, not theory. Apprentices learn by doing real packs from day one. Schools and bootcamps publish curricula as forkable libraries.
For society
Preserve trades and crafts before they fade. Lower the barrier to entry for new workers and entrepreneurs. Make the practical knowledge of humanity legible — and accessible to all.
Built like open source
The standard belongs to everyone. So does the library.
OpenPSL works the same way open source software does. You can fork any public Procedure Pack, adapt it to your context, and optionally contribute improvements back. Industry communities can curate their own libraries. Translations are welcomed. The standard is MIT-licensed; community packs default to CC-BY-SA.
Fork freely
Take any public pack as a starting point. Adapt it. Republish or keep private — your call.
Publish back
If your improvements help others, set the pack to public. The community library grows with every contributor.
Preserve trades and crafts
From formal trades to informal practices, capture knowledge that would otherwise live only in one head. Every domain matters.
Open governance
The schema evolves through open discussion. Breaking changes are versioned; backward compatibility is a goal, not an aspiration.
Roadmap
Where this is going.
OpenPSL is an open project, evolving in the open. The roadmap is a direction, not a contract.
- Now
Open standard + reference platform
Universal JSON schema with 12 fields. Visual builder with drag-and-drop steps, multi-role assignment everywhere, structured time estimates, four kinds of decision rules (nested if/then up to 4 levels, hierarchical matrices, scientific-formula expressions, fallbacks), a 3×3 risk heatmap, control-to-risk linking, public/private packs, forks, versions, exports to JSON / Markdown / PDF, OAuth sign-in with Google, Microsoft and GitHub, and a community seed library across the most automatable industries.
- Next
Multilingual UI · Pack linking · Public API
Localized interfaces in major languages. Procedures that reference other procedures (composability). A public API so external systems and agents can read and write packs.
- Later
Execution engine · BPMN bridges · Verified packs
Run a procedure end-to-end, tracking who did what when. Bridges to BPMN, ERP, ticketing systems. Verified industry packs with certified maintainers. Marketplaces for premium libraries.
- Vision
The operational layer for the agent economy
Any AI agent — yours, ours, or a competitor's — can read any OpenPSL pack and act on it. Procedures become as portable as web pages. Operational knowledge becomes a global commons.
The best procedure is the one that exists.
Pick one piece of work you do. Capture it in five minutes. Save it, fork it later, share it if it helps someone. That's how a global library of human know-how gets built — one pack at a time.