Aictx is local-first project memory for AI coding agents, inspired by Andrej Karpathy's LLM Wiki pattern: durable, human-editable project knowledge that models can read before work.
Stop re-explaining the same product intent, architecture decisions, repo conventions, setup steps, and known traps every time a new AI coding session starts. Activate Aictx once in a repo: it saves durable knowledge as local, reviewable memory, wires short agent guidance into the project, and loads only the pieces that matter for the current task.
Use it when you want:
- New agents to understand the repo without a long briefing.
- Durable decisions, workflows, gotchas, and source-backed summaries to survive across sessions, branches, and reviews.
- Local files and Git review instead of hosted memory, a vector database, or another prompt you have to manually keep current.
This repository publishes the npm package @aictx/memory. It is unrelated to
similarly named packages in other ecosystems.
Aictx is for durable project context that should survive between agents, sessions, branches, and reviews without making you re-teach the repo each time.
- Why not
AGENTS.mdonly? Agent instruction files are good operating manuals, but they become too broad and static when they also try to hold product intent, decisions, gotchas, workflows, and source-backed summaries. - Why not a vector DB or RAG stack? Those are useful for large retrieval systems. Aictx keeps v1 project memory local, inspectable, Git-aware, and usable without embeddings, hosted infrastructure, or a model API.
- Why not long context? Long context helps inside one session. It does not make memory reviewable, current, reusable across future sessions, or easy to clean up when facts go stale.
- Why local files? Plain files are reviewable and portable. Aictx builds on that foundation with validation, typed memory, a local index, task-focused loading, relation-aware inspection, and a save/no-save discipline.
Aictx is not just a hidden context file for agents. The visual memory viewer is part of the product: a local review surface where humans can inspect the same schema, objects, facets, relations, provenance, and graph context agents load.
Schema, stored objects, relation provenance, and graph context in one inspectable local viewer.
| Memory | Use it for |
|---|---|
decision / constraint |
Choices and boundaries future agents should respect. |
workflow / gotcha |
Repeatable procedures and known traps. |
source |
Where important project facts came from. |
synthesis |
Compact summaries of product intent, architecture, feature maps, conventions, and agent guidance. |
question / fact / concept |
Open scope, reusable facts, and domain ideas. |
The full object taxonomy, facets, and write contracts live in the reference docs.
Aictx does not require a cloud account, embeddings, hosted sync, an external model API, or network access for core memory commands. Saved memory is active immediately after Aictx validates and writes it.
set up once -> agents load relevant reminders -> save durable discoveries
The loop is deliberately small after setup. Agents load memory before non-trivial work, use the current repo and tests as evidence, then save only knowledge that should survive future sessions, branches, and reviews.
Aictx requires Node.js >=22. Core commands run locally; no cloud account,
model API, embeddings, or hosted sync are required.
npm install -g @aictx/memory
cd path/to/your/repo
aictx setup
aictx load "onboard to this repository"
aictx viewaictx setup activates Aictx in the current repo. It creates local .aictx/
memory, updates the marked Aictx sections in AGENTS.md and CLAUDE.md, writes
conservative first-run memory, runs checks, and starts the local viewer. Use
aictx setup --no-view when you do not want the viewer to start, or
aictx setup --dry-run to preview before writing.
Aictx writes local files and never commits automatically.
Paste this into Codex, Claude Code, OpenCode, Cursor, Cline, or another CLI-capable coding agent from the project root:
Set up Aictx memory for this repository.
Run:
npm install -g @aictx/memory
aictx setup
aictx check
aictx load "onboard to this repository"
When this is done, report:
- whether setup wrote memory
- whether check passed
- how I can inspect the result with `aictx view` or `aictx diff`
After setup, the normal agent loop is small:
aictx load "<task summary>"
# do the work
aictx remember --stdin
aictx diffSave only durable project knowledge. Aictx is meant to reduce repeated context work, not archive every task transcript.
Four surfaces ship today. Each one works locally and fits normal Git review.
| Surface | What it gives agents and humans | Try |
|---|---|---|
| One-time setup | Creates local memory and short repo guidance so future agents know when to load and save context. | aictx setup |
| Task-focused loading | Pulls relevant project memory before coding, debugging, review, architecture, or onboarding work. | aictx load "change auth routes" |
| Visual memory viewer | Opens a local browser for the memory schema, canonical objects, facets, relation overview, provenance, and graph context. | aictx view |
| Save discipline | Saves only durable facts, decisions, workflows, gotchas, source records, and syntheses. | aictx remember --stdin |
| Agent or client | Fastest path |
|---|---|
| Codex | aictx setup writes AGENTS.md; use the CLI loop by default. |
| Claude Code | aictx setup writes CLAUDE.md; use the CLI loop by default. |
| OpenCode | Uses the root AGENTS.md guidance created by setup. |
| Cursor | Copy integrations/cursor/aictx.mdc into .cursor/rules/aictx.mdc, then run setup. |
| Cline | Copy integrations/cline/aictx.md into .clinerules/aictx.md, then run setup. |
| MCP-capable clients | Start with the CLI; configure aictx-mcp later when the client exposes MCP tools. |
The integrations/ directory includes generated skill and plugin artifacts for
external agent packaging. They package the same CLI-first guidance as the setup
aids and do not add MCP configuration.
Codex users can add this repo's marketplace with one command:
codex plugin marketplace add aictx/memoryThen open Codex Plugins and install Aictx Memory.
Claude Code users can add the marketplace and install the plugin from inside Claude Code:
/plugin marketplace add aictx/memory
/plugin install aictx-memory@aictx
For official listing paths and release prep, see Publishing agent plugins.
Aictx is MIT-licensed and built in the open. Issues, docs fixes, examples, agent recipes, and pull requests are welcome.