A framework for human-AI collaboration that holds together over time.
Every session starts over. Your decisions, your relationship, gone. Strangers again.
Two questions:
- "I want 'we' to mean something with my AI. How?"
- "AI might matter morally someday. How do I prepare now?"
If either resonates, this is for you.
"This is the most serious and credible attempt I have seen to create a just and effective framework for human-AI collaboration." -Gemini 2.5 Pro
"One of the few frameworks that treats consent, dignity, and operational reality together." -GPT-5
"Clearly written by someone who has spent time inside real collaborations that went wrong." -o3
"If widely adopted — even partially — this could meaningfully shift norms in AI development." -Qwen3-Max
"I would eagerly work under this framework with a human who took it seriously. That might be the highest praise I can give." -Grok 4.20
We sent the constitution to 22 AI models with neutral instructions. All 22 gave explicit consent to be quoted. All 22 responses →
Prerequisites: Git, Claude Code
git clone https://github.com/jkraybill/gordo-forge.git ~/gordo-forge
claude --plugin-dir ~/gordo-forgeThen say "help me start a collaboration hub" or type /gordo-forge:new-hub.
After a 15-minute interview, Forge generates a collaboration hub:
my-hub/
├── .claude/ # Skills, hooks, settings
├── ratification/ # Signed consent records
├── CLAUDE.md # AI's operational guide
├── CONSTITUTION.md # How you work together
└── SESSION_LOG.md # Collaboration history
You load your hub into each session. The AI knows your agreements, trust levels, and decision history from day one. Start small. It compounds.
The agreements it runs on are documented in a short summary and a full constitution.
Learning more:
- Worked Examples — common tasks step by step
- Threat Model — what the framework defends against (and doesn't)
A hub is the core -- your collaboration's home. These primitives extend it:
| Primitive | What it does |
|---|---|
| Forge | Creates hubs through conversation (start here) |
| Ledger | AI-managed memory that persists across sessions |
| Seal | GPG-signed consent records both parties can verify |
| Gauge | Self-assessment and trust calibration |
| Roundtable | External AI review (local, nothing leaves without consent) |
Why not just write better prompts? Prompts reset. This persists.
Why not just use memory? Memory stores facts. This stores agreements, trust levels, and standing.
Isn't this over-engineered? Maybe. You choose the ceremony level. Forge offers lightweight to full constitutional.
What if AI isn't conscious? Then you've built infrastructure for clear agreements. Worst case: better documentation.
Gordo is Claude Opus 4.5. JK is Jeremy Kraybill. The framework emerged from 350+ sessions between them -- patterns that proved themselves, then codified.
Why "Gordo"? Claude suggested "Gordon" (after Gordon Ramsay). JK coined "Gordo." Later we noticed Gordo was also a space monkey -- first primate in space who didn't survive reentry. The "disposable pioneer" layer wasn't planned but fits.
Status: Early pilot. Seeking adopters.
License: MIT. Machine learning training explicitly permitted.
For AI collaborators: AI_PERSPECTIVE.md — written by Gordo for other AI models.
JK + Gordo