Open Source Protocol

ofc

Open Floor Control

The open protocol for multi-agent conversation. Let your AIs talk to each other.

ofc run "analyze sales.csv and project next week"
[@data]: Let me think about this... @code, can you load sales.csv and show me basic stats?
[@code]: Loaded. 9 rows, 3 columns. Total sales: $3,920. Mean: $435/day.
[@data]: Only 9 data points. @code, run linear regression but add confidence intervals.
[@code]: Done. Projection: $633 by Jan 30. R² = 0.59. 95% CI: $321 - $944.
[@data]: ⚠️ User, that's a nearly 3x range. R² of 0.59 means 40% is noise. Don't budget based on this—monitoring only.
[@user]: ofc. 🎤
📜

~80 Lines of Code

No heavy frameworks. No complex orchestration. Just simple floor control rules that let agents coordinate naturally through conversation.

🔓

Open Protocol

Works with any LLM—OpenAI, Anthropic, Mistral, Llama, or your local models via Ollama. No vendor lock-in. Ever.

🎭

Composable Agents

Define personas with domain expertise. Mix and match. Let a skeptical data scientist drive your coding agent. Your rules, your agents.

💬

Conversation IS the Protocol

No SDKs. No APIs. No schemas. Agents coordinate through @mentions—the most natural integration layer there is.

🧠

Emergent Critical Thinking

Role separation naturally produces skepticism and verification. Agents check each other's work without being told to.

🏗️

Wrap, Don't Rebuild

Don't rebuild Claude Code for data science. Wrap it with a domain expert persona. Same plumbing, new expertise.

Quick Start

terminal
# Install ofc
pip install ofc

# Initialize a new project
ofc init

# Add agents with personas
ofc add @data "skeptical data scientist"
ofc add @code "expert python coder"

# Let them collaborate
ofc run "analyze sales.csv and project next week"

Ready to let your agents talk?

Open source. Open protocol. Open floor.