metadata
title: cluas_huginn
emoji: 💬
colorFrom: yellow
colorTo: purple
sdk: docker
pinned: false
hf_oauth: true
hf_oauth_scopes:
- inference-api
license: apache-2.0
short_description: A gathering of guides, a council of counsels
🐦⬛ cluas huginn (thought's ear) 🐦⬛
- A Multi-Agent Research Council
A gathering of guides, a council of counsels
What It Does
Four specialised AI agents with distinct epistemic roles debate questions using structured dialectic, building knowledge over time.
Thesis: Present findings with evidence
Antithesis: Challenge, verify, and provide counterpoints
Synthesis: Build consensus and update shared memory
Evolution: Future discussions build on accumulated knowledge
Why It's Different
- Distinct epistemic roles: Each agent uses specialized tools
- Structured dialectic: Reduces bias, promotes steelmanning
- Persistent memory: Knowledge accumulates over time
- Tool-personality mapping: Raven verifies news, Corvus searches literature, etc.
The Council
| Agent | Role | Icon |
|---|---|---|
| Corvus | Academic verifier (literature) | 🐦⬛ |
| Raven | Accountability enforcer (news) | 🦅 |
| Magpie | Trend explorer (patterns) | 🪶 |
| Crow | Grounded observer (environment) | 🕊️ |
Modes
- Collaborative Mode: Ask a question and receive synthesised research with sources
- Active Mode: Join the discussion, steer research, challenge claims
Dialectic Process
- Thesis: Characters present initial findings using specialized tools
- Antithesis: Characters challenge, verify, and provide counter-evidence
- Synthesis: Council builds consensus and adds to collective memory
- Evolution: Future discussions build on accumulated knowledge
Why It Matters
Most AI assistants are stateless. cluas_huginn Council remembers, learns, and builds knowledge over time.
🛠 Tech Stack
- Base: Python, Gradio
- LLMs: UserKeys/Groq/Nebius/etc (various models, llama3.3 main)
- Tools: Academic search, news verification, web exploration
- Memory: Persistent character memories
Taglines
- "A gathering of guides, a council of counsels"
- "Research that remembers, knowledge that accumulates"
- "Multi-agent MCP research collective"
Architecture
- Corvus: Academic verifier (literature search)
- Raven: Accountability enforcer (news verification)
- Magpie: Trend explorer (pattern connector)
- Crow: Grounded observer (environmental data)
Key Innovations
- Unified inheritance architecture
- Shared epistemic principles with "Four Temperament" character differentiation
- Tool-use heuristics per character
- Steelmanning and collaborative disagreement built-in
What Makes This Different
- Not just multiple LLMs - distinct epistemic roles
- Structured dialectic (thesis/antithesis/synthesis)
- Tool usage guided by character personality
- Collaborative, not adversarial
This project is licensed under the Apache 2.0 License.