OdinAn agent that runs the work, not just the words.

Self-hosted execution agent for Discord — shell and SSH, browser, schedules, sub-agents, knowledge, and a management WebUI, all inside limits you set.

Install

Architecture

GitHub

⌘

67 built-in tools

Shell and SSH on registered hosts, context-checked patches, background processes, browser automation, web, scheduling, knowledge, memory, and sub-agents.

Tool reference

⧉

Sub-agents and loops

Spawn isolated agents, message them mid-run, wait for snapshots, and read complete results by cursor. Autonomous loops with explicit limits.

How results are delivered

⏱

Scheduled operations

Cron and one-time schedules, webhook-triggered workflows, delegated background tasks — output posted back to the originating channel.

Configuration

☰

Management WebUI

Live execution viewer, agents, processes, schedules, audit, sessions, usage, hosts, config, turn state — 211 REST routes behind it.

API reference

⊘

Bounded by policy

Permission tiers, per-user host access, a command governor, secret redaction, and an audit log with optional HMAC tamper evidence. Odin executes what it is asked, inside what it is allowed.

Security model

⇄

Three model backends

OpenAI Codex over a ChatGPT subscription (GPT-6 and GPT-5.x, multi-account), Kimi, or local Ollama. Switch at runtime from the WebUI.

Provider setup

74built-in tools · 23 core

211REST routes, order-pinned

11,712tests · 346 files

3model backends

MITlicense · Debian package

operator

the staging API is returning 502s since the deploy — find out why and fix it if it's safe

odin

Probed the health endpoint → 502 in 40 ms, upstream refused. The service unit has been flapping since 14:02; the journal ends in address already in use :8080 — the previous worker was never reaped. Killed the orphan, restarted the unit, health is 200 in 12 ms across five probes. Validation bundle passed 4/4: service active, port open, HTTP 200, no new errors in 60 s. Audit record written.

An illustrative exchange — hostnames invented, behaviour real. Tool calls are recorded in the audit log and shown live in the WebUI.

Odin WebUI — Capabilities → Tools: every tool, how often it has run, and a switch to remove it from the model

WebUI · Capabilities → Tools. Every built-in tool, its usage on this install, and a per-tool switch that removes it from the model's catalog.