Persistent Reasoning Graphs
Every thinking session becomes a queryable graph with explicit edges for influence, support, contradiction, and refinement.
Open-Source Research Platform
AI reasoning you can see, challenge, and audit.
Opus Nx transforms Claude's extended thinking into persistent reasoning graphs you can navigate, verify, and steer. Built on Anthropic's mission to create reliable, interpretable, and steerable AI systems.
ThinkFork Analysis
Fork hard problems into competing reasoning styles, steer branches in flight, and inspect convergence over time.
Metacognitive Insights
Generate structured audits of bias, patterns, and improvement hypotheses from the model's own reasoning traces.
Platform Capabilities
- 6-agent swarm orchestration
- Persistent session memory and artifacts
- Research-first evaluation loops
Quick Start
- 1. Clone the repo and run
./scripts/docker-start.sh - 2. Add your
ANTHROPIC_API_KEY - 3. Open localhost:3000 and start reasoning.
Next.js 16Python 3.12SupabaseClaude Opus 4.6TypeScript
Aligned with Anthropic's Mission
Reliable. Interpretable. Steerable.
Every feature in Opus Nx maps to a core principle of safe AI. We don't just use AI — we make its reasoning transparent, verifiable, and controllable.
Reliable
Verified reasoning you can trust
Process Reward Model scores every reasoning step. The 6-agent swarm cross-validates conclusions. Nothing is accepted on faith — every claim is checked.
- PRM step-by-step verification
- Multi-agent cross-validation
- Geometric mean confidence scoring
Interpretable
See every reasoning step
Extended thinking becomes a persistent, navigable graph. Decision points are explicit nodes. Metacognitive audits surface hidden biases and patterns.
- ThinkGraph persistent reasoning
- Explicit decision point nodes
- Metacognitive bias detection
Steerable
Direct reasoning in real-time
Fork reasoning into competing styles and steer branches mid-flight. Human-in-the-loop checkpoints let you course-correct at any decision point.
- ThinkFork 4-style branching
- Mid-flight branch steering
- Human-in-the-loop checkpoints
Research Foundation
Grounded in Peer-Reviewed Research
Tree of Thoughts
ThinkForkYao et al., 2023
BFS/DFS search over reasoning trees with state evaluation
Let's Verify Step by Step
PRM VerifierLightman et al., 2023
Process supervision — verify each reasoning step independently
Graph of Thoughts
GoT EngineBesta et al., 2023
Arbitrary thought graph topology with aggregation and refinement
MemGPT
Memory HierarchyPacker et al., 2023
3-tier memory hierarchy with paging and auto-eviction
Open Source
Built in public. Contribute to the
future of AI reasoning.
Opus Nx is MIT-licensed and community-driven. Run it locally with your own keys, extend the reasoning engines, or contribute to the agent swarm.