What is Precursor?

Opinionated approach to work follow-up, built as an AI assistant.

Precursor is an AI assistant for following up on work — the threads, meetings and tasks that unfold over days or weeks rather than in a single sitting.

It's built from several distinct surfaces, each suited to a different mode of working: long-lived topics that can hydrate themselves from a GitHub issue, throwaway chats for a fast answer, recorded live meetings, autonomous agents for long-running tasks, and workflows that chain those agents into a reusable pipeline. They share one set of tools, personas and memory, so what you teach it in one place applies everywhere.

Under the hood Precursor is deliberately compact: normal runtime is a single uvicorn worker that serves a JSON API and the pre-built React SPA from the same process.

Why it exists

Most AI chat tools are a flat list of disconnected conversations. Work isn't flat — it's organized around issues, meetings, and tasks with their own lifespan. So instead of one undifferentiated stream, Precursor gives each kind of work a surface that fits it, and keeps the context attached to the work rather than to the conversation you happened to have about it.

A topic linked to a GitHub issue rebuilds its context from that issue on every turn, so newer comments outweigh older ones and the assistant always reasons over the current state. An agent keeps a durable scratchpad across runs. A workflow remembers what it already processed. In each case the work carries its own memory.

Highlights

• Topic-scoped, tree-organized conversations, each optionally linked to a GitHub issue whose labels tag the chat — alongside quick chats for throwaway questions.
• Streaming chat over Server-Sent Events (sse) with markdown, mermaid, and code highlighting.
• Bring your own model — GitHub Copilot (default), Azure AI Foundry, or any OpenAI-compatible gateway.
• MCP both ways — Precursor exposes its conversations and features as an MCP server and attaches external MCP tool servers to extend its capabilities.
• Live meeting assistant — Meeting transcription via Azure Speech, live insights, Q&A, and an editable summary.
• Agents mode (opt-in) — hand a complex task to an autonomous Copilot SDK agent and follow it in a workflow-style timeline.
• Workflows (opt-in) — chain those agents into a reusable, named pipeline (research → draft → review) that runs unattended on a schedule or a webhook, with quality gates and human approval checkpoints.
• Skills & memory, a scheduler with recurrence and guards, reminders, workspaces the assistant can edit, and a Kanban board over your issues.
• Plugin-ready — backend entry points plus a frontend extension registry.

The stack in one line

Python 3.12+ · FastAPI · SQLAlchemy 2 (async) · Alembic · Vite + React 19 + TypeScript · Tailwind CSS · the mcp and openai SDKs — with uv as the one toolchain for env, run, build, and release. See the technical stack for the full breakdown.

Where to next

• Installation — get the app running locally.
• Quick start — your first topic in a couple of minutes.
• Configuration — GitHub token, model providers, and settings.
• Feature guides — a tour of everything Precursor can do.

Single-user by design

Precursor ships with no authentication and is meant to run bound to 127.0.0.1. Don't expose it to a network without your own authenticating reverse proxy in front of it. See the security model.