Beacon is an engineering intelligence layer that sits above GitHub, CI/CD, coding agents and team communication. It continuously ingests what happens across an engineering organisation into a single append-only event stream, then derives status, dashboards and answers from that stream instead of asking people for updates.

Software engineering intelligence (SEI) is the practice of measuring engineering work from the systems it already produces — version control, CI, issue trackers, agent runs — rather than from self-reported status. An SEI platform normalises those signals into a shared model so leaders can see delivery, flow and risk without instrumenting people directly.

Jira and Linear are systems of record: a human moves a ticket, and the board reflects what that human remembered to do. Beacon stores no status at all. Work item status is a projection folded from the event stream — task.started resolves to in progress, pr.opened to in review, pr.merged to done — so the board cannot drift from what the repository actually shows.

Coding agents and CI push structured events straight to Beacon over a single authenticated endpoint, POST /api/events, using a workspace API key. An agent reports agent.planning when it forms a plan, agent.tests_passed when a suite goes green, and agent.completed when work is verified done. Identity resolution attributes each event to the right engineer and the right work item, so agent output appears in the same stream as human output rather than in a separate tool.

Beacon takes signals two ways. Connectors pull from tracked sources — GitHub repositories today — and normalise them into typed events such as code.commit, pr.merged and task.status_changed. The ingestion API lets coding agents, CI pipelines and internal scripts push events directly. Notes, documents and meeting summaries can also be ingested into a semantic knowledge base that the AI chat searches alongside the event stream.

Yes. Beacon exposes a Model Context Protocol server at /api/mcp with OAuth dynamic client registration, so MCP-capable coding agents such as Claude Code, Codex and Cursor can authorise themselves and then read work items, plans, documents, events and the knowledge base as first-class tools. The agent writing the code and the system reporting on it talk to the same API.

By folding the events attached to it, in order. Beacon never writes a status field that a projection would later contradict; status is recomputed from the stream, which means the history that produced it is always inspectable. If a task shows as done, the merge commit that made it done is one click away.

No, and the role model enforces it rather than merely promising it. Engineers see their own teams’ timeline, work and chat, and roster names only — no peer metrics. Managers see dashboards for the teams they are responsible for. Team lead is a per-team flag granting manager-like visibility for that one team. The design rule is to assist engineers, not to monitor them.

Beacon is invite-only. An admin issues a one-time /join link that carries a predefined role and team assignment and expires after seven days; sign-in is GitHub OAuth against the account named on that invite. A GitHub account with no workspace membership is refused at the callback.