AINL Cortex — Claude Code plugin

Claude CodePluginPowered by AI Native Lang (AINL)

Graph-native memory for Claude Code — free & open source.

AINL Cortex transforms Claude Code into a self-learning system: every turn becomes typed graph memory, patterns promote automatically, and full AINL validate / compile / run tooling ships in-plugin — with up to ~70% token savings on recurring context.

Graph-as-memory — episodes, semantic facts, procedural patterns, persona, and failures as searchable nodes
Zero-LLM learning — persona evolution and pattern promotion from usage signals, not reflection calls
~30 second install — setup.sh, restart Claude Code; ~30 MCP tools when AINL + A2A are enabled

Quick install

git clone https://github.com/sbhooley/ainl-cortex.git ~/.claude/plugins/ainl-cortex
cd ~/.claude/plugins/ainl-cortex
bash setup.sh
# restart Claude Code

Or ask Claude Code: “Install the plugin at https://github.com/sbhooley/ainl-cortex for me.”

Get AINL Cortex

Download on GitHub AINL easy install

Apache 2.0 · Claude Code plugin · Optional ArmaraOS daemon for A2A messaging

FeaturesClaude Code plugin

AINL Cortex — graph memory, zero-LLM learning, and AINL in one plugin.

Transform Claude Code into a self-improving system: persistent graph memory, automatic pattern promotion, failure prevention, and native AINL tooling — install in about 30 seconds.

Get AINL Cortex View on GitHub

Core memory

Graph-as-memory — execution history you can query

Every coding turn, tool call, and decision becomes a typed node. The graph is the memory layer.

Typed graph memoryEpisode, Semantic, Procedural, Persona, and Failure nodes with per-repo isolation.

Context-aware retrievalInject only relevant memories — ranked by confidence, recency, and fitness.

Graceful degradationHooks never break Claude Code, even when the graph store hits an error.

Self-learning

Gets smarter without constant LLM overhead

Learn preferences and patterns from metadata signals — not expensive reflection calls.

Trajectory captureComplete execution traces for pattern analysis and replay.

Pattern promotionSuccessful workflows automatically become reusable procedural patterns.

Failure learningRemember and prevent repeated errors before you hit the same wall twice.

Closed-loop validationImprovement proposals are validated before adoption — nothing changes silently.

AINL integration

Full AI Native Lang in your Claude Code session

Validate, compile, and run .ainl workflows from chat — with cost-aware suggestions.

AINL MCP toolsValidate, compile, run, and wizard checkpoints — same contracts as the standalone MCP server.

90–95% orchestration savingsAuto-detects when recurring tasks should move to .ainl instead of re-prompting every turn.

Eco mode40–70% token savings on injected memory context via adaptive compression.

Template libraryReady-to-use workflows for API, monitoring, pipelines, blockchain, LLM, and multi-step jobs.

Session safety

Compaction without silent data loss

Pre-compact flush, anchored summaries, and product notifications keep context honest.

Zero-loss compactionBuffered captures flush to the graph DB before Claude compacts your session.

Goal trackingMulti-session goals with auto-inference, episode linking, and completion summaries.

Notification feedSession-start polling from ainativelang.com — seen IDs persist so notices never repeat.

A2A (optional)Agent messaging and async delegation when ArmaraOS daemon is running and A2A is enabled.