Petri

github.com/onthemarkdata/petri

Grow your own AI context.

An agent orchestration framework that grows your AI's context. Petri decomposes claims into DAGs of logical units and validates them bottom-up through a multi-agent adversarial review pipeline.

The result is a repository of curated information, citations, and URLs — organized by concepts, how they relate to each other, and the nuances of the claims. Useful as pre-loaded context assets for AI agents.

What It Does

  • Decomposes a claim into a DAG of logical units (cells)
  • Runs each cell through 13 agents across Socratic analysis, research, critique, debate, red team, and evaluation phases
  • Curates citations and URLs alongside the reasoning trail
  • Ships with both a CLI for AI agents and an interactive UI for reviewing active agents

Built For

  • AI agents that benefit from pre-loaded, citation-backed context
  • Researchers stress-testing claims through adversarial review
  • Engineers building agentic systems that need durable reference material
1 # Petri
2
3 [github.com/onthemarkdata/petri](https://github.com/onthemarkdata/petri)
4
5 **Grow your own AI context.**
6
7 An agent orchestration framework that grows your AI's context. Petri decomposes claims into DAGs of logical units and validates them bottom-up through a multi-agent adversarial review pipeline.
8
9 The result is a repository of curated information, citations, and URLs — organized by concepts, how they relate to each other, and the nuances of the claims. Useful as pre-loaded context assets for AI agents.
10
11 ## What It Does
12
13 - Decomposes a claim into a DAG of logical units (cells)
14 - Runs each cell through 13 agents across Socratic analysis, research, critique, debate, red team, and evaluation phases
15 - Curates citations and URLs alongside the reasoning trail
16 - Ships with both a CLI for AI agents and an interactive UI for reviewing active agents
17
18 ## Built For
19
20 - AI agents that benefit from pre-loaded, citation-backed context
21 - Researchers stress-testing claims through adversarial review
22 - Engineers building agentic systems that need durable reference material

No editor is open

Open a file from the Explorer or use Ctrl+P

TERMINAL
Welcome to markfreeman.dev terminal
Type 'help' for available commands.
 
visitor@markfreeman.dev:~$