13 Patent-Protected Innovations | Apache 2.0

Your Development
Environment That Thinks

CogniGraph transforms any codebase into a self-governing reasoning network. Install it, run one command, and your Claude Code sessions become governed, self-improving, and semantically validated.

Get Started Free View on GitHub

# Install

pip install cognigraph[api]

# Initialize your project

cd my-project

kogni init

# Scans repo, builds KG, creates CLAUDE.md

# Injects GCC + GSD + Ralph Loop protocols

# Registers MCP server for Claude Code

# Open Claude Code → governed development ready

Innovations

0.757

Constrained F1

99.7%

Governance Accuracy

Cloud Infrastructure

How It Works

From install to governed development in 60 seconds

pip install One command

→

kogni init Scans repo, builds KG

→

Claude Code MCP tools available

→

Graph Reasoning Agents think together

→

Learning Gets smarter each session

MCP Tools for Claude Code

7 governed tools in every session

Each tool is powered by graph-of-agents reasoning over your project's knowledge graph.

kogni_context FREE

Smart context loading for session start. Returns relevant KG nodes, active branch, and applicable lessons in ~300 tokens. Replaces reading large flat files.

kogni_reason FREE

Graph-of-agents reasoning over your project KG. Each node becomes an agent that reasons locally, exchanges messages, and produces synthesized answers.

kogni_inspect FREE

Inspect your project knowledge graph. Explore nodes, edges, statistics, and relationships between components.

kogni_preflight PRO

Governance preflight check before code changes. Returns relevant lessons, past mistakes, applicable ADRs, and safety boundary warnings.

kogni_lessons PRO

Find lessons and past mistakes relevant to your current operation. Lessons accumulate over time and auto-promote by severity.

kogni_impact PRO

Trace downstream impacts through the dependency graph. See which components are affected and their risk level before you deploy.

kogni_learn PRO

Feed outcomes back for Bayesian graph learning. Success strengthens edges, failure weakens them. Your KG gets smarter every session.

Development Protocols

Structured governance that no competitor has

Three battle-tested protocols injected into your CLAUDE.md. Always free.

GCC

Global Context Controller

Session continuity across conversations. Branch management. Auto-commits every 30 minutes. Never lose context again.

Session start/end protocols
Structured COMMIT checkpoints
Branch management (BRANCH, MERGE)
Context retrieval at 6 granularity levels
Multi-instance coordination

GSD

Get Shit Done

Structured pre-work before execution. Spec before code. Context engineering over brute force. Atomic commits.

DISCUSS — clarify scope and constraints
PLAN — atomic tasks with dependencies
EXECUTE — one task = one commit
VERIFY — 3-source cross-reference check

RALPH

Autonomous Iteration Loop

Persistent autonomous iteration with safety guards. Feed task + binary criteria, work until done or blocked.

Binary completion criteria (pass/fail)
Max 20 iterations (configurable)
Cannot deploy to main
Cannot perform destructive ops
Scope creep = auto-exit

Research

13 patent-protected innovations

European Patent Application EP26162901.8 | Published paper with benchmarks

#	Innovation	Module	Tier
1	PCST Activation — sublinear subgraph selection	activation.pcst	FREE
2	MasterObserver — zero-cost transparency layer	orchestration.observer	FREE
3	Convergent Message Passing — similarity termination	orchestration.convergence	FREE
4	Backend Fallback Chain — heterogeneous inference	backends.fallback	FREE
5	Hierarchical Aggregation — topology-aware synthesis	orchestration.aggregation	FREE
6	SemanticSHACLGate — 3-layer OWL governance	ontology.semantic_shacl_gate	PRO
7	Constrained F1 — joint quality + governance metric	benchmarks.constrained_f1	PRO
8	OntologyGenerator — automated OWL+SHACL	ontology.generator	PRO
9	Adaptive Activation — dynamic Kmax	activation.adaptive	PRO
10	Online Graph Learning — Bayesian edge updates	learning.graph_learner	PRO
11	LoRA Auto-Selection — per-entity adapters	adapters.auto_select	PRO
12	TAMR+ Connector — retrieval-to-reasoning	connectors.tamr	PRO
13	MCP Plugin — governed context engineering	plugins.mcp_server	FREE

Your Models, Your Keys

Bring your own API backend

Zero vendor lock-in. Zero cloud infrastructure from us. You control the models and the costs.

Anthropic

Claude Haiku / Sonnet / Opus
Recommended for best results

OpenAI

GPT-4o / GPT-4o-mini
Full compatibility

AWS Bedrock

Any Bedrock model
EU region support

Custom Endpoint

Any OpenAI-compatible API
Self-hosted models

Pricing

Start free. Scale when ready.

The free tier is genuinely powerful. Pro unlocks the full moat.

Free

$0 /forever

Full SDK + GCC/GSD/Ralph protocols. Powerful enough for most developers.

✓ Innovations 1-5 (core reasoning)
✓ 3 MCP tools (context, inspect, reason)
✓ GCC + GSD + Ralph Loop protocols
✓ Repo scanning + KG building
✓ Single backend
✓ CLAUDE.md injection
✗ Advanced MCP tools
✗ Graph learning

Pro

$29 /dev/month

Full moat. 13 innovations. Self-improving knowledge graph.

✓ Everything in Free
✓ All 13 innovations
✓ All 7 MCP tools
✓ Semantic SHACL governance
✓ Online graph learning
✓ Adaptive activation
✓ Multi-backend fallback
✓ Session analytics

Team

$99 /team/month

Shared governance for teams up to 10 developers.

✓ Everything in Pro
✓ Shared knowledge graph sync
✓ Cross-developer lessons
✓ Multi-instance coordination
✓ Custom domain ontologies
✓ Team analytics dashboard

Enterprise

Custom

Private deployment. Compliance reporting. SLA support.

✓ Everything in Team
✓ Private / air-gapped deployment
✓ Compliance reporting (SOC 2, ISO)
✓ Custom SHACL governance rules
✓ Dedicated support + SLA
✓ Custom integrations

Your DevelopmentEnvironment That Thinks