3 commands. That's it.

pip install graqle[api]
graq init
graq reason "what breaks if I change auth?"

3 nodes activated. 500 tokens. 5 seconds. $0.0003.

Not 60 files. Not 50,000 tokens. Not $0.15. Not guessing.

What Graqle does

Your AI assistant reads files one at a time. It doesn't understand your architecture — it reads text.

Graqle builds a knowledge graph of your entire codebase. Every module becomes a reasoning agent. When you ask a question, only the relevant nodes activate, reason about their domain, and synthesize one answer.

pip install graqle → graq init → Knowledge Graph
                                       │
             ┌──────────┬──────────────┼──────────┐
             ▼          ▼              ▼          ▼
           CLI      Python SDK     MCP Server  Studio UI
        (terminal)  (scripts)      (any IDE)   (dashboard)

The graph is the product. Your machine, your API keys, your data.

What's new in v0.26.0

Intelligence Compilation

Graqle now compiles your knowledge graph into actionable intelligence:

graq compile                    # Compile intelligence from your graph

This produces a .graqle/intelligence/ directory with:

• 396 module packets — risk scores, impact radius, consumers, dependencies
• 135 insights — warnings, suggestions, connections, superlatives
• Risk heatmap data — LOW/MEDIUM/HIGH/CRITICAL per module
• CLAUDE.md auto-injection — your AI assistant learns your architecture automatically

Governance Gate (DRACE)

Every reasoning session is scored on 5 axes: Data quality, Relevance, Accuracy, Completeness, Evidence strength.

graq verify                     # Pre-commit governance check

• Hash-chained audit trails (tamper-evident)
• Evidence chains linking decisions to source
• Scope gates preventing out-of-domain reasoning
• Auto-recompile when intelligence goes stale

Multi-Signal Activation (Neo4j)

Gate + Rerank: semantic score gates everything, topology signals only amplify.

final = semantic × (1 + authority + memory + link + freshness)

Max amplification: 1.45×. An irrelevant node can never sneak through.

Self-Learning Loop

graq learn → graph grows → git commit → auto-recompile → CLAUDE.md updates → AI gets smarter

The graph remembers which nodes produce useful answers. Gets better with every query.

Studio Dashboards

6 new visual dashboards: Intelligence, Governance, Health, Learning, Control, Share.

Works with everything

No cloud account. No infrastructure. No config files to write.

CLI

# Reasoning
graq reason "what depends on auth?"          # Graph reasoning
graq context auth-lambda                      # 500-token focused context
graq inspect --stats                          # Graph statistics
graq "what is safe to refactor?"              # Natural language (auto-routed)

# Build & compile
graq init                                     # Scan repo, build graph, wire IDE
graq scan repo .                              # Rescan codebase
graq scan all .                               # Code + JSON + documents
graq compile                                  # Compile intelligence layer
graq verify                                   # Governance gate + staleness check

# Teach
graq learn node "auth-service" --type SERVICE
graq learn edge "Payments" "auth" -r DEPENDS_ON
graq learn discover --from "auth-service"     # Auto-discover connections
graq learn doc architecture.pdf               # Document ingestion

# Studio & server
graq studio                                   # Launch visual dashboard
graq serve                                    # Start REST API
graq doctor                                   # Health check

Python SDK

from graqle.core.graph import Graqle
from graqle.backends.api import AnthropicBackend

graph = Graqle.from_json("graqle.json")
graph.set_default_backend(
    AnthropicBackend(model="claude-sonnet-4-6")
)

result = graph.reason(
    "What services depend on auth?",
    max_rounds=3,
    strategy="top_k"
)

print(result.answer)
print(f"Confidence: {result.confidence:.0%}")
print(f"Cost: ${result.cost_usd:.4f}")

MCP Tools

Available automatically in Claude Code, Cursor, VS Code, and Windsurf after graq init:

Studio Dashboard

graq studio                        # Opens http://127.0.0.1:8888/studio/

14 backends, one config line

# graqle.yaml
model:
  backend: groq    # or: anthropic, openai, bedrock, gemini, ollama, deepseek, mistral, ...

Task-based routing

Different models for different tasks. Fast models for lookups, smart models for reasoning:

routing:
  default_provider: groq
  rules:
    - task: reason
      provider: anthropic
      model: claude-sonnet-4-6
    - task: context
      provider: groq
      model: llama-3.1-8b-instant

Auto-scaling backend

Graqle starts with JSON/NetworkX (zero deps). When your graph grows:

pip install graqle[neo4j]     # Adds neo4j driver — same API, just faster

Document-aware intelligence

Graqle connects code to documents to configs in one graph.

Auto-linking: exact match, fuzzy match, semantic match, LLM-assisted. Cross-source deduplication. Contradiction detection.

The numbers

Installation

pip install graqle              # Minimal (no API keys needed)
pip install graqle[api]         # + Anthropic, OpenAI, Bedrock
pip install graqle[docs]        # + PDF, DOCX, PPTX, XLSX
pip install graqle[neo4j]       # + Neo4j graph database
pip install graqle[all]         # Everything

15 Innovations (Patent EP26162901.8)

Every innovation is free under Apache 2.0.

Quick start

# Python project
pip install graqle[api] && cd my-project && graq init
graq reason "what depends on the database module?"

# With documents
pip install graqle[api,docs]
graq scan all . && graq scan wait
graq reason "what does the architecture doc say about auth?"

# Claude Code (auto-wired)
graq init    # Claude Code now has graq_context, graq_reason, graq_impact, etc.

# Cursor
graq init --ide cursor

Pricing

Free for individuals. Always.

Citation

@article{kumar2026graqle,
  title   = {Graqle: Governed Intelligence through Graph-of-Agents Reasoning
             over Knowledge Graph Topologies with Semantic SHACL Validation},
  author  = {Kumar, Harish},
  year    = {2026},
  institution = {Quantamix Solutions B.V.},
  note    = {European Patent Application EP26162901.8},
  url     = {https://github.com/quantamixsol/graqle}
}

Contributing

git clone https://github.com/quantamixsol/graqle
cd graqle
pip install -e ".[dev]"
pytest                         # 2,000+ tests

See CONTRIBUTING.md for development setup, testing, and PR guidelines.

License

Apache 2.0 — use it commercially, modify it freely, keep the attribution.