Private AI Research System
FRA
Feather Research Agent
FRA is a private, self-hosted AI research system. It is not a commercial product, not a cloud service, and does not compete with ChatGPT, Claude, or any other AI platform.
FRA is two things working together: a console for talking to any AI model you choose, and an autonomous research agent that works while you sleep. Both run on your own computer. Both are governed by real mathematical theorems — not marketing claims.
You choose the engine. DeepSeek. Claude. Ollama. Any model on your machine or any API you have a key for. FRA is the workspace. The intelligence comes from whatever you connect.
Two Components
FRA Sesh
The Console — Sesh means “scribe” in the language of ancient Kemet
Chat with any AI model. Run agents with tool access. Deep research. Document editing. Email triage. Calendar. Notes. All in one private workspace you own.
Built on Odysseus (MIT licensed). One command to install on Windows, Mac, or Linux.
FRA Rekh
The Research Agent — Rekh means “to know” in the language of ancient Kemet
Autonomous research intelligence. Runs in the background. Multi-agent hierarchy with fault isolation. Persistent memory. Evidence tracking. Self-auditing.
Built on Agent Zero (MIT licensed). Runs any model — DeepSeek, Claude, OpenAI, Ollama, local models.
Mathematical Governance
What makes FRA different is not the AI models it uses — it’s how it governs them. Every decision passes through a mathematical framework derived from proved theorems:
| Feather Completion Theorem | Finds the natural missing piece, rejects invented patches, classifies every path |
| SBT (Stability-Balance Theorem) | Detects when the agent is drifting from optimal — stops overclaiming before it happens |
| Vine Theorem | Optimal sub-agent count using exact strand cardinalities (52/21/6, 295/120/36/8) |
| Ordered Chamber | Enforces valid operating domains — tasks outside capacity are rejected |
| Cauchy Ledger | Triple-entry audit trail — every agent action recorded as source + flow – loss |
| Ma’at Evidence Taxonomy | No claim exceeds its evidence level — assumption, runtime, test, verified |
| 67+ Lean4 Verified Theorems | SBT programme formally verified — lake build clean, 0 errors, 0 sorry |
| K4,n Jensen Hyperbolicity | Degree-4 Jensen polynomials of Riemann xi are hyperbolic for all n — proved |
These are not research papers. They are the operating system. Every module in FRA is a theorem in code form.
How to Get FRA
FRA is open source. MIT licensed. You run it on your own computer.
# Windows — one command
git clone https://github.com/pewdiepie-archdaemon/odysseus.git
cd odysseus
py -3.11 -m venv venv
venvScriptsActivate.ps1
pip install -r requirements.txt
python setup.py
python -m uvicorn app:app –host 127.0.0.1 –port 7000
Open http://localhost:7000. Log in. Configure your models in Settings. DeepSeek, Claude, Ollama, anything with an API key or running locally.
Why FRA Exists
Most AI tools ask you to trust a company. FRA doesn’t. You see every line of code. You control every model. You own every piece of data.
FRA was built for one person — its creator, Sterling Hayden. It grew from a simple need: a private workspace where AI could help with real research, real mathematics, real decisions — without sending anything to anyone else’s servers.
If that sounds useful, it’s yours. No sign-up. No subscription. No cloud.
Ma’at is the measure. The feather does not lie.
FRA is open source software. MIT License. Credits: FRA Sesh built on Odysseus by pewdiepie-archdaemon. FRA Rekh built on Agent Zero by frdel. The Feather Framework, Completion Theorem, SBT, Vine Theorem, and all mathematical governance systems are original work by Sterling Dudley Hayden.