Orca

About the project

Orca is a tool for managing autonomous AI coding agents. It was built around a practical problem: one agent in a terminal is useful, but several agents working on the same project quickly need planning, state, oversight and safety boundaries.

The goal is not to replace the developer. The goal is to give agents an operating frame: what they should do, in what order, with what level of autonomy, and when the work should stop for a human decision.

What it solves

Agentic development often creates the same kind of operational noise:

• one terminal is running, another is waiting, a third is stuck,
• useful output disappears inside scrollback,
• dependencies between tasks have to be tracked manually,
• sensitive work needs approval,
• the human does not know whether the system is progressing or idle.

Orca adds a control layer over that. Tasks have state, missions have capacity, agents run in isolated sessions, and the dashboard shows what is happening right now.

Architecture

Orca is designed as a standalone daemon with an API, CLI and web interface. The daemon owns task state, missions, sessions and activity events. Agents are launched in isolated tmux sessions and report results back through a unified CLI callback.

The main layers are:

• daemon and API — task, mission, session, event stream and configuration management,
• web UI — dashboard, kanban, timeline, missions, projects and settings,
• CLI — fast terminal workflow and callback layer for agents,
• agent runtime — launching Claude Code, Codex, OpenCode or other executors,
• guardrails — blocking risky areas and controlling autonomy.

Tech stack

RUNTIME   → Node.js 22 + TypeScript
API       → Hono + REST + SSE
DATA      → SQLite + WAL
WEB       → Next.js 16 + React 19 + Tailwind 4
EDITOR    → Monaco
AGENTS    → tmux + CLI executors
CLI       → native Node CLI
TESTS     → Vitest

A key design detail is that the daemon does not need to parse every agent’s stdout format. Agents have one clear way to close work and submit a result back to the system. That keeps the integration surface clean and extensible.

Key capabilities

• Missions and phases — a larger goal can be split into ordered steps with capacity and status.
• Controlled autonomy — different tasks can run with different levels of independence.
• Live sessions — running agents are visible in the UI, including terminal streams.
• Activity timeline — events, states and signals remain traceable.
• Project editor — the web interface can browse project files, diffs and commits.
• Open-source base — the project is designed as a local-first tool without requiring an external service.

Who it is for

Orca is useful for developers and teams that already use AI agents regularly and are hitting the limits of manual coordination. It fits workflows where one prompt is not enough and the work has several steps: audit, implementation, tests, review, fixes and documentation.

It is not a generic task manager. It is an operating layer for agentic development.

Outcome

The project shows what the next step of AI-assisted development can look like. Not one smarter chatbot, but a system that can launch multiple agents, track their state and keep humans in control.

For the portfolio, Orca is a product-engineering case study at the intersection of AI, developer tools, systems design and operational UI. It combines an agent runtime, real terminal processes, a web dashboard, safety decisions and open-source distribution into one product.