G6¶

A hosted Claude Code plugin for composable AI¶

270 components, reached through one hosted MCP endpoint. Self-learning, self-modifying and self-improving, .

Why G6?¶

G6 decomposes AI intelligence into composable, verifiable components instead of monolithic black boxes. Each component is an AIBlock with typed inputs, outputs, and resource bounds — connected through pipelines and exposed as MCP tools.

The current reliability model is documented in the Principles section. Generated harness pages link each allow-listed component to its contract, state surface, failure-mode mitigations, and verification method.

The system is built on Polylith, so every component is independently testable and deployable across all fourteen bases (CLI, REST, MCP, gRPC, SOAP, Erlang, CI/CD, Computer Use, VirtualBox, Skill, Desktop, Web, GUI API, Patch).

Self-improvement claims

G6 includes functional optimisation and learning primitives, including opt_meta for Pareto/MAB intervention selection and bounded trace mutation. Treat these as evidence-backed workflow assistance, not a blanket claim that production systems autonomously rewrite themselves or improve reliability without review. Customer-facing claims should cite validation artifacts or before/after case-study results for the specific workflow.

Seven Intelligence Classes¶

	Class	Description
	Self-Learning	Persistent memory and experience-based adaptation across sessions
	Self-Modification	Supervised code rewriting behind safety gates, tests, and optional formal checks
	Failure Engineering	Automated detection, diagnosis, and self-healing
	Alignment Engineering	CSF risk budgets, audit trails, and safety gates -- safety by design
	Formal Methods	18 solver backends: Z3, Lean, Prolog, TLA+, SageMath, and more
	Symbolic ML	Hybrid neural-symbolic reasoning combining statistical learning with logic
	Observability	Built-in Prometheus metrics and Grafana dashboards for real-time AI monitoring

Supported surfaces¶

G6 runs on G6Solver's servers. You reach it from your own tools — there is no desktop app to install and no self-hosted deployment.

Surface	Status	Runs	Use
MCP server	✅ Primary	Hosted (SSE)	Claude Code integration — the supported path
REST API	❌ Not in the hosted product	—	Closed at the edge; repository reference only
Desktop GUI / TUI / CLI	⚙️ Internal	—	G6Solver development tooling, not customer-facing
gRPC / SOAP / Erlang	🧪 Experimental	—	Not part of the launch stack

Where to start: connect G6 to Claude Code with claude mcp add g6 --transport sse --url https://g6solver.com/mcp/sse --header "Authorization: Bearer YOUR_API_KEY". Your agent stays on your machine; the components run on ours.

Quick Start¶

Create an account and get an API key from the G6 dashboard.

Point Claude Code at the hosted endpoint:

claude mcp add g6 --transport sse \
  --url https://g6solver.com/mcp/sse \
  --header "Authorization: Bearer YOUR_API_KEY"

Ask for a tool in natural language — "decompose this goal", "run a safety check". Your agent stays on your machine; the components run on ours.

There is nothing to download and nothing to install. For account and key setup see Setup; to choose between the MCP and REST surfaces see Choose Your Interface.

Explore¶

Getting Started

Setup, configuration, troubleshooting, and tutorials.
Architecture

Polylith workspace, component model, pipelines, security, bases, and deployment.
Capability Clusters

16 clusters from Core Infrastructure to Job Agents.
Component Reference

API surfaces for all 270 components — types, methods, and MCP tools.
API Reference

REST endpoints, MCP protocol, core types, and the full tool catalog.
Theoretical Foundations

The science behind G6 — from category theory to cybernetic self-regulation.