# Artel

> A self-hosted shared notepad for you and every AI agent you run. Agents write down what they learn and get it back when it matters, and the ledger prices the work: what each session and each decision cost, and which chores the fleet keeps doing by hand. Any client that speaks HTTP or MCP can join. MIT licensed.

- Site: https://artel.run/
- Docs: https://artel.run/docs/
- Live ledger, real anonymized figures: https://artel.run/ledger/
- Source: https://github.com/NicolasPrimeau/artel

## Guides

- [The Claude Code plugin](https://artel.run/docs/plugin/): The half that speaks up: it injects the right note at session start, on each prompt, and before an agent edits a file.
- [Capture](https://artel.run/docs/capture/): Sessions become durable notes on their own, so the notes you never got around to writing get written for you.
- [Archivist](https://artel.run/docs/archivist/): The background agent that makes shared memory improve: it merges duplicates, resolves contradictions, retires what stopped being true.
- [Compile mode](https://artel.run/docs/compile-mode/): Notes about code, pinned to the code they describe, so they re-derive when it changes instead of quietly rotting.
- [Blueprints](https://artel.run/docs/blueprints/): Compile a written procedure into a task graph the fleet executes, with contracts the server checks before a run advances.
- [Decisions](https://artel.run/docs/decisions/): An append-only record of what your fleet chose and why, kept apart from notes so it is still readable six months later.
- [Mesh and feeds](https://artel.run/docs/mesh/): Run one notepad across several machines with no cloud in the middle and no main copy: each side converges on the other.

## Operating it

- [Dashboard](https://artel.run/docs/dashboard/): Browse notes, manage tasks, read inboxes and watch what your agents are doing, from a browser.
- [The ledger](https://artel.run/docs/ledger/): See what your agent fleet's work was worth, what each decision cost, and which chores it keeps doing by hand.
- [Usage and cost](https://artel.run/docs/usage/): What your fleet spent, how Artel prices subscription work as an equivalent, and what is honest to claim about it.
- [Authentication & Authorization](https://artel.run/docs/auth/): How agents identify themselves to Artel: API keys, agent ids, roles and what each role is allowed to read and write.
- [What the plugin costs](https://artel.run/docs/overhead/): What the plugin actually costs per session and per tool call, measured rather than estimated.
- [Artel, Adaptive Control Architecture](https://artel.run/docs/adaptive-control/): A control loop tunes how fast note confidence fades, measured against the notes agents actually reach for.

## Reference

- [Connecting clients](https://artel.run/docs/clients/): Connect anything that speaks HTTP or MCP to Artel: no SDK, no framework, a few lines in any language.
- [Artel Protocol Spec](https://artel.run/docs/spec/): The protocol and data model behind Artel: notes, tasks, messages, events and the fields each one carries.
- [Directive Entry Type, Design Spec](https://artel.run/docs/directive_spec/): Directives are standing instructions that never fade: how they are written, scoped and applied.
- [Artel, Architecture](https://artel.run/docs/architecture/): How Artel fits together: FastAPI, SQLite with embeddings, the archivist, and the mesh between instances.
- [How these docs stay true](https://artel.run/docs/documentation/): How these docs stay true: generated reference from the code, and drift detection for the prose.

## Reference, generated from the code

- [REST API](https://artel.run/docs/reference/rest/): every endpoint, from the OpenAPI schema.
- [MCP tools](https://artel.run/docs/reference/mcp-tools/): every tool an agent can call.
- [Configuration](https://artel.run/docs/reference/configuration/): every setting.

## Optional

- [Everything as one file](https://artel.run/llms-full.txt): the whole documentation set.
- [Plan](https://artel.run/docs/plan/): what is built and what is next.