A.I.M. — Claude Code Edition

"Treat your AI like a bot, not an oracle."

A.I.M. is an open-source exoskeleton designed to cure the "Amnesia Problem" of autonomous AI agents. It wraps around Claude Code and forces it to act like a disciplined Principal Engineer — externalizing memory into a local SQLite Engram DB, automating the Git lifecycle, and ruthlessly distilling your context the exact second a coding session ends.

Core Features

• The Single-Shot Memory Compiler: Instantly distills session context and surgically edits MEMORY.md the second a coding session ends. No background cron jobs. No silent LLM hallucinations while you sleep.
• The Zero-Token Python Engine (The Autonomic Layer): Deterministic Python scripts strip raw terminal JSONL noise into a clean "Signal Skeleton," reducing token weight by up to 85% before an LLM ever touches the data.
• The Federated Brain (Archipelago Model): Eliminates database bottlenecks by segregating memory across purpose-built SQLite databases (project_core.db, global_skills.db, datajack_library.db, subagent_ephemeral.db).
• Strict GitOps Bridge (Atomic Deployments): Forces agents to use aim-claude bug, aim-claude fix <id>, aim-claude push, and aim-claude promote. Agents are physically forbidden from raw git commit or pushing to main, ensuring every change is isolated in a Worktree and test-driven.
• The Obsidian Bridge & Remote Fleet: Sync your project folder to an Obsidian Vault and let a headless A.I.M. daemon on a secondary GPU server compile your memory, giving you an infinitely scalable, cross-machine brain.
• The DataJack Protocol (.engram Cartridges): Package thousands of pages of documentation into pre-vectorized .engram files. Run aim-claude jack-in framework.engram to instantly inject semantic recall of entire libraries without spending a single API token.
• Modular Cognitive Routing: Decouple the "conscious" and "subconscious." Keep flagship models (Claude Opus/Sonnet) in the terminal for coding, and route background tasks (memory indexing) to free local models via Ollama.
• Context Collapse Shield (Failsafe Snapshot): A rolling dead-man's switch continuously saves your last 10 turns. If the agent crashes, the context is automatically salvaged.
• Executive Guardrails (Anti-Drift): The cognitive_mantra hook tracks autonomous tool calls. At 50 actions, it forcefully halts execution and forces the agent to recite its GitOps rules, washing away context degradation.
• Reincarnation & Crash Recovery: Run aim-claude reincarnate to perform an automated context handoff to a fresh agent, or aim-claude crash to salvage an interrupted session and resume exactly where you left off.
• Strict Bug Reporting: The aim-claude bug command strictly requires explicit --context, --failure, and --intent flags to ensure the next "blind" agent inherits full epistemic certainty.
• Interactive TUI Cockpit: A visual terminal interface (aim-claude tui) to configure LLM routing, guardrails, and context limits.
• Universal IDE Support (MCP): A built-in MCP server exposes the Engram DB to any connected IDE (Cursor, VS Code, Claude Desktop) without platform-specific adapters.

---

All comprehensive documentation, architectural maps, setup instructions, and origin stories live in the wiki. The code is the source of truth; the wiki is the map.

READ THE OFFICIAL WIKI

---

The A.I.M. Ecosystem

DISCLAIMER: WORK IN PROGRESS The repositories below are experimental adaptations. The aim repository is the primary "Soul" of the project. The core architectural decisions, the Engram DB logic, and the central integrations happen there first before being ported to the external adaptations.

• aim: The Core Engine (Built for Gemini CLI).
• aim-claude: Adaptation for Anthropic's Claude Code. (You are here)
• aim-codex: Adaptation for OpenAI's GPT Codex.
• aim-antigravity: The experimental GUI/MCP Desktop adaptation.

Buy Me a Coffee