You’re paying for the package, not the chat.
Pick the SME bundle you walk out with from $19.
Three artifact tiers, ranked by what every modern LLM expects. Free ships the 3-file starter you can paste into Claude or ChatGPT. Paid ($19 for one Twin or $49/mo for unlimited) ships the full SMETP v0.5 portable package— Anthropic-style SKILL.md, references, per-LLM compiled wrappers, validation manifest, monitors. Ultimate adds calibrated confidence and on-prem.
Three markdowns: skill.md, workflow.md, graph.md. Drop them into Claude, ChatGPT, Gemini, Perplexity, MCP. No card, no signup, no lock-in.
Anthropic-style SKILL.md, references/ (skill-document.json, decision-logic, validation manifest, provenance, edge-cases), providers/ (Claude system prompt, ChatGPT instructions, Gemini system, OpenAI tools, MCP server), monitors/, scripts/. The folder modern agents expect.
Paid plus a fitted confidence_model scaffold, group-fairness checks, adversarial pack, drift thresholds tuned for your domain (PSI / ECE / override-rate). Ships as MIT-licensed open source — bring your own LLM and storage, keep every byte inside your VPC.
Free
Walk out with a real SME bundle in 5 minutes. No card. No signup.
You walk out with: Three markdowns — skill.md, workflow.md, graph.md — that you can paste into any LLM.
The 3-file starter every LLM accepts.
- skill.md — drop into Claude, ChatGPT, Gemini as a system prompt
- workflow.md — the decision flow your agent loop replays
- graph.md — entities + a mermaid diagram for Notion / GitHub
- Live preview as you talk — copy or download anytime
- Up to 40 discovery turns per SME
- Side-by-side test: vanilla LLM vs your bundle
- Refresh = it’s gone (intentional)
Paid · One Twin
The full v0.5 portable package. Every LLM. Every runtime. One zip.
You walk out with: The full SMETP v0.5 portable skill package — SKILL.md + references/ + per-LLM compiled wrappers + monitors. As a zip.
Save one SME forever, with the bundle modern agents expect.
- Anthropic-style thin SKILL.md (progressive disclosure)
- references/skill-document.json (canonical wire schema)
- references/decision-logic.md, dictionary.md, edge-cases.md, policy-sources.md
- references/elicitation-provenance.md (CDM/ACTA sessions, kappa)
- references/validation-manifest.json (SME-match, outcome-match, Brier, ECE)
- providers/ — claude system prompt, chatgpt instructions, gemini system, openai-tools, mcp-server.json
- monitors/thresholds.yaml + scripts/execute.py runtime stub
- Hosted /twin/<id> chat URL · public or private
- 1,000 Twin API calls / month included (then $5 / 1k)
Paid · Studio
For teams shipping many SME packages — unlimited Twins, shared workspace.
You walk out with: Same full v0.5 portable package as Twin — but unlimited Twins, team workspace, shared library.
399/yr saves ~32%.
- Everything in Paid · One Twin, unlimited
- Team workspace + shared package library
- One-click deploy to Claude · ChatGPT · Cursor · MCP
- Voice interviews (ElevenLabs) included
- 25,000 Twin API calls / seat / month
- Priority compile queue · custom ontology packs
Self-host
Run the whole protocol on your own infrastructure.
You walk out with: The full SMETP stack as MIT-licensed open source. Run @smetp/runtime on your own infra, bring your own LLM, keep every byte in your VPC. No license cost, no lock-in, ever.
Bring your own LLM. Bring your own storage. Keep your data.
- Everything in Paid · Studio, self-hosted
- MIT-licensed @smetp/spec + sdk + cli + runtime
- Bring your own LLM (Anthropic / OpenAI / local)
- Knowledge graph in your VPC or on-prem
- No telemetry · no phone-home
- Open ontology pack format (write your own)
- Reference Docker compose + Helm chart
- Issues + PRs welcome on GitHub
Every saved Twin ships with a measurable trust contract — workflow coverage, cited-evidence rate, deterministic ratio, latency.
The compiled SkillDocument is portable: Anthropic Skill, OpenAI tools, Gemini function-calling, MCP — or self-host with @smetp/runtime.
Every Twin you save sharpens the next one — your dictionary, your ontology pack, your decision patterns become reusable across the team.
What’s in the package, side-by-side
Free ships 3 markdowns. Paid ships the full SMETP v0.5 zip. Ultimate ships paid + the calibration / monitoring layer the literature requires.
| Feature | Free | Paid · Twin | Paid · Studio | Ultimate |
|---|---|---|---|---|
| Inside the package | ||||
| skill.md · workflow.md · graph.md (3-file starter) | ✓ | ✓ | ✓ | ✓ |
| SKILL.md (Anthropic-style thin nav, frontmatter, when-to-use) | — | ✓ | ✓ | ✓ |
| MANIFEST.yaml · README.md · CHANGELOG.md | — | ✓ | ✓ | ✓ |
| references/skill-document.json (canonical wire schema) | — | ✓ | ✓ | ✓ |
| references/decision-logic · dictionary · edge-cases · policy-sources | — | ✓ | ✓ | ✓ |
| references/elicitation-provenance.md (CDM/ACTA, kappa) | — | ✓ | ✓ | ✓ |
| references/validation-manifest.json (SME-match · outcome-match · Brier · ECE) | — | ✓ | ✓ | ✓ |
| providers/claude · chatgpt · gemini · openai-tools · mcp-server | — | ✓ | ✓ | ✓ |
| monitors/thresholds.yaml + scripts/execute.py runtime stub | — | ✓ | ✓ | ✓ |
| Calibrated confidence_model (isotonic / Platt fit) | — | — | — | ✓ |
| Adversarial pack + group-fairness checks | — | — | — | ✓ |
| Drift thresholds tuned per domain (PSI / ECE / override-rate) | — | — | — | ✓ |
| Capture & iterate | ||||
| Discovery turns per SME | 40 | Unlimited | Unlimited | Unlimited |
| Live preview · copy / download | ✓ | ✓ | ✓ | ✓ |
| Side-by-side test vs vanilla LLM | ✓ | ✓ | ✓ | ✓ |
| Voice interviews (browser-native, premium voice on paid) | browser | ✓ | ✓ | ✓ |
| Persisted Twin saved to your account | — | ✓ | ✓ | ✓ |
| Hosted /twin/<id> chat | — | ✓ | ✓ | ✓ |
| Number of Twins you can keep | 0 (ephemeral) | 1 per purchase | Unlimited | Unlimited |
| One-click deploy to Claude / GPT / Cursor / MCP | — | — | ✓ | ✓ |
| Team workspace + shared library | — | — | ✓ | ✓ |
| Custom ontology packs (upload your own) | — | — | ✓ | ✓ |
| Governance | ||||
| SSO · audit · on-prem / VPC (run it yourself) | — | — | — | ✓ |
| MIT license · bring your own LLM and storage | — | — | — | ✓ |
| Issues + PRs welcome on GitHub | — | — | — | ✓ |
Common questions
What’s the difference between the Free, Paid and Ultimate packages?▼
Free is three markdowns (skill.md, workflow.md, graph.md) you paste into any LLM as a system prompt. Paid (Twin or Studio) is the SMETP v0.5 portable package as a zip — Anthropic-style SKILL.md plus references/ (skill-document.json, decision-logic, validation manifest, provenance), providers/ (Claude system prompt, ChatGPT instructions, Gemini system, OpenAI tools, MCP server), monitors/, and a runtime stub. Ultimate adds a calibrated confidence_model, adversarial pack, group-fairness, drift thresholds and on-prem hosting.
Why can any LLM use these packages?▼
Because that’s exactly what the package layout is designed for. SKILL.md follows the Anthropic Agent Skills convention (frontmatter + when-to-use + when-not-to-use + progressive disclosure to references/). The providers/ folder ships a pre-compiled wrapper for each major host: a Claude system prompt, ChatGPT Custom GPT instructions, a Gemini system_instruction JSON, OpenAI function-calling tool definitions, and an MCP server descriptor. Drop the right file in, the LLM behaves like your SME.
Why $19 for one Twin?▼
Because saving your first SME package should feel like buying a coffee, not a SaaS. We'd rather you buy ten packages for the price of one "Pro plan" and tell ten friends. Upgrade to Studio happens naturally when you realize you have a 4th, 5th, 8th SME to capture.
What happens if I refresh during the free flow?▼
It’s gone. That’s intentional — it makes the difference between "trying it" and "owning it" obvious. Download the 3-file starter before you close the tab and you still walk away with real files you can drop into Claude or ChatGPT.
Why Studio when I can keep buying Twins for $19?▼
Studio gives you unlimited Twins (same paid package), a team workspace, one-click deploy to Claude / ChatGPT / Cursor / MCP, voice included, and 25k Twin API calls per seat per month. Most operators break even on Studio at the third package in a month.
Can I run SMETP entirely on my own infra?▼
Yes. The @smetp/spec, @smetp/sdk, @smetp/cli, and @smetp/runtime npm packages are MIT-licensed and bring-your-own-LLM. The hosted product adds the live capture canvas, the v0.5 package compiler, the validation runner, and the multi-tenant knowledge graph.
What does the Self-host tier include?▼
The whole SMETP stack as MIT-licensed open source — @smetp/spec, @smetp/sdk, @smetp/cli, @smetp/runtime — published on npm and GitHub. Bring your own LLM, your own storage, run it inside your VPC. No license fee, no telemetry, no lock-in. The hosted Twin/Studio tiers are for people who don't want to run infrastructure; self-host is for people who do.
SMETP is open source · MIT-licensed
Read the method. Run it yourself. No license cost, no lock-in.
The Subject Matter Expertise Transfer Protocol is a practical way to capture expertise once and let any modern LLM stack reuse it. The hosted tiers above are a convenience — everything you need to run the whole pipeline yourself ships as MIT-licensed npm packages.