Orbit — Linear for AI-Native Developers
The Orbit dashboard (orbit web serve) — task backlog, live audit log, per-agent scoreboard.
Orbit is durable, intent-tracked agentic project management for developers who use AI coding agents heavily — Linear / Jira designed for the AI-native solo developer, with a path to team-scale automation as trust in agents matures.
The wedge today is the individual engineer driving multiple coding agents (Claude Code, Cursor, Aider, Codex CLI) against real code, who has outgrown the one-agent-one-terminal model. You need a durable backlog when ideas and bugs accumulate faster than any single session can hold. You need lifecycle tracking when work spans sessions, weeks, or branches. You need intent attribution when six months from now you have to remember why an agent wrote a given line of code. Linear and Jira solve durable project management for human-driven teams. Agent vendors solve in-session execution. Orbit is the layer above — the layer that turns individual agent sessions into a coherent body of work.
Built for the AI-native solo developer first, with a deliberate funnel that expands toward teams over time: solo adoption → internal champion at a team → team-scale agentic automation. The team-scale destination is multi-year. The wedge is today.
The full positioning — what Orbit is for, who it's for in funnel order, what it refuses to become, the open-core commercial model, and the commercial roadmap for the hosted Team product — lives in docs/POSITIONING.md. Read it before contributing or evaluating fit.
Primary Features
Three features carry the wedge. Everything else is infrastructure that makes these reliable.
Durable, intent-tracked task layer — available today
The wedge surface. Every task carries a durable lifecycle (proposed → backlog → in-progress → review → done) that survives across sessions, branches, and weeks. Every commit produced through Orbit carries the task_id, so the codebase itself becomes a queryable record of agent intent — git log --grep '\[T20260506-...\]' finds every commit that flowed through a given task; orbit task show reconstructs the prompt, plan, execution trace, and review threads for that task months later.
The hand-tracking-in-Notion-or-scratch-notes problem ends here. Ideas and bugs go into orbit task add, get worked on by an agent (now or later), and produce a body of work that's still navigable a year from now. This is what Linear and Jira solve for human-driven teams; Orbit solves it for the AI-native solo developer who plans to expand the practice across their team.
Knowledge-graph–aware tooling — available today
Agents inspect a parsed, content-addressed graph of your codebase: directories, files, extracted symbols, import edges, crate dependencies, trait implementors, and signature-matched caller/reference indexes. Queries return token-budgeted packs shaped for prompt consumption, not LSP-style hover text.
The graph is the technical moat and it is measured, not asserted. Under MCP exposure, the graph reduces token cost on structural code questions (see benchmarks/graph/ for full v3 results: codex hybrid arm came in 35% cheaper than no-graph, with pre-registered cull thresholds passed). It is also branch-scoped and built for safe parallel execution — two worktrees on two branches rebuild concurrently without corruption; reads fall back to the default branch until a new branch has been built. The public graph surface is read-only; write coordination happens before dispatch through task context_files and orbit.task.locks.reserve preflight guards.
Design docs: docs/design/knowledge-graph/.
Auditability — available today
Every tool call, provider request/response, and task-state transition is a structured, queryable event with agent identity attached. When something goes sideways — a bad merge, a regression, a refactor whose intent you can't reconstruct months later — you answer what / why / who without asking the maintainers. Append-only, tamper-evident, exportable.
For the wedge audience (single engineer, multiple agents, real code), this is what makes ad-hoc agent sessions trustworthy enough to live with downstream — both for you, and for the future teammate who reviews your PR. For the long-arc team-scale destination, it's also the substrate every team-grade primitive depends on.
Full contract below in the Auditability section. Design docs: docs/design/auditability/.
The Core Model
Orbit is built around four concepts you will actually touch:
