# Stele — agentic ledger

> A knowledge graph and agent memory for AI coding agents — Claude Code,
> Cursor, Codex, and any MCP-compatible client. Tasks, decisions, lessons,
> and risks live in one shared record an agent walks before it answers, so
> every prompt starts with what the project already learned.

Stele is one of: (1) a task system, (2) a knowledge graph of decisions /
lessons / risks / commitments, and (3) an operator that drives agentic work
through those tasks. The point is that all three share a single ledger — so
each part makes the others sharper.

Today it runs as a Claude Code plugin (MCP server + hooks + slash commands)
against a hosted Supabase backend. The web UI at https://app.stele-ai.dev
lets humans read the record. The CLI (`stele`) handles auth, admin, and a
local web server.

Status: v1 in beta. Free during beta. Sign in with GitHub.

## Core

- [Landing page (full markdown)](https://stele-ai.dev/llms-full.txt): The complete content of the marketing landing — hero, in-practice demo description, the real refresh-lock incident replay, the six capabilities, install snapshot, the four surfaces (Claude Code, MCP clients, Web, CLI), and where it's going.
- [App](https://app.stele-ai.dev): The hosted web UI — tasks, knowledge, docs, the embedded assistant.
- [GitHub](https://github.com/serkanyersen/stele): Source. Single dev, stealth-ish but public repo.

## What it does

- **Risks surfaced before you reintroduce them** — reachable risk nodes ride along on the context trail; past incidents named before the change repeats them.
- **Doctor keeps the ledger curated** — `stele doctor` audits the graph as it grows; near-duplicates surfaced for dedupe, broken edges caught.
- **Pushes back on bad changes** — when a request contradicts a prior decision, the agent can decline and cite the decision instead of quietly redoing the work.
- **Provenance you can follow** — design doc → task → issue → decision → person who made the call.
- **Fewer repeated mistakes** — agents check changes against what the project already learned.
- **Kept honest** — entries carry provenance and age; stale knowledge is flagged and revised.

## Optional

- [Concept: the agentic ledger](https://stele-ai.dev/llms-full.txt#concept): Why a shared record matters more when agents are doing the writing.
- [Concept: one backend, four surfaces](https://stele-ai.dev/llms-full.txt#surfaces): Supabase + RLS as the entire backend; thin clients for agents and humans.