Agents Remember Documentation
Drift-aware repository memory for coding agents in complex codebases. Captures what code can’t say on its own! Retrieves memory by path, semantics, and relationsship (code-graph).
Core Features
Agents Remember gives coding agents project memory they can verify and act on. It turns local invariants, naming rules, migration scars, cross-repo contracts, and “this looks safe but is not” facts into versioned Markdown beside the code, checks that memory against Git before use, and updates it only after approved work lands.
src/orchestrator/core_editor.py
ar-memory/onboarding/src/orchestrator/core_editor.py.md
- Path-addressed memory: A source file’s note lives at a deterministic mirror path, so an agent holding a file can reach the right context without search, ranking, or guesswork.
- Git-proven freshness: File notes, route overviews, and entity catalogs are drift-checked against source commits, route scopes, or deterministic fingerprints before they are trusted.
- Search that finds, not decides: Optional semantic memory and code-graph providers help locate relevant files, callers, dependencies, and concepts, but verified Markdown and source code remain the truth.
- Memory that lands with code: External memory repos use a
memory.md ledger, isolated dual worktrees, preview/apply closeout, and all-or-nothing integration so code and memory stay synchronized.
- Repo-owned agent behavior: Each memory repo carries
system/ files for path rules, tools, coding guidelines, documentation sources, branch policy, and reporting shape, so the same project rules load across harnesses.
- Harness-ready first run: Starter packages for Claude Code, Codex, Cursor, Antigravity, VS Code Copilot, Hermes, Pi.dev, and OpenClaw carry the native MCP, skills, hooks, rules, and instruction files each harness needs.
The default setup stores durable memory in the target repository under ar-memory/. Teams that need separate memory repositories can use external memory under ar-coordination/memory-repos/ar-<repo>/. For the full tour, see Features.
Start Here
- Getting Started - the agent-driven setup path: copy the harness starter package, wire the MCP with
uvx, restart once, then let c-13-install-and-onboard run runtime_install and set up memory, onboarding, and providers.
- Features - the concentrated product tour: by-path memory, drift, providers, worktrees, harness setup, and operational guardrails.
- Concepts - the memory model and the terms used throughout the project.
- Architecture - how the source checkout, installed runtime, coordination root, and memory roots fit together.
- Workflows - the
l-01-session-job-lifecycle lifecycle, its build modes, and worktrees.
- Benchmark Methodology - paired benchmark runs, metrics, validity checks, and limitations.
- FAQ - design principles, objections, and comparisons.
Install Guides
Guides
Reference