The AI coding agent built for the terminal.

Available as a terminal interface, desktop app, or IDE extension. Supports 75+ LLM providers, including local models.

Binary releases and package channels arrive with 0.1.0.

Built into hordecode

Every number below is drawn from the live, audited source — see the docs for sources.

17 Harness Pipeline skills plan, TDD, debug, review, verify
18 Research Pipeline skills scoped to research-only permissions
15 Guardrails permission rules, stuck-agent detection, sandboxed workflows
11 Memory operations hybrid search, entity tracking, dedup
[*]Supervisor

A persistent session that routes work across a fleet of background agents, approves permissions, and audits every result before reporting back — you're never left babysitting subagents.

[*]Harness Pipeline

Every task runs through the matching skill before implementation starts — 17 skills covering plan, TDD, debug, review, verify — with a hard bar: no completion without a verification step actually run.

[*]Research Pipeline

A dedicated pipeline for investigation and deep research, separate from the build pipeline — 18 of its own skills, scoped to research-only permissions.

[*]Live agent visibility

Every active agent, subagent, and cron job shows up in the sidebar in real time, with status and elapsed time — full visibility into what's actually running, not a black box.

[*]Guardrails

15 durable, database-persisted safety mechanisms — permission rules, idle- and stuck-agent detection, a sandboxed workflow engine — built into the runtime, not bolted on after the fact.

[*]Hybrid RAG

11 memory operations — hybrid keyword/semantic search, entity tracking, session summaries, dedup, all built in. No plugin, no separate API key, no third-party service to trust.

hordecode

Terminal UI screenshot, in hordecode's own color theme — coming once the TUI redesign ships.