Odin Agent Runtime

A CLI agent that obeys your governance.

Lives inside the workspace. Loads only the skills your tenant enabled. Uses your provider keys. Caps spend by construction. Audits every spawn to BrainDB.

Request enterprise access See the workspace

BuilderOperator

Composition

15 proven-core skills + your tenant's opt-ins.

Every Odin Agent Runtime image carries 15 proven-core skills that load regardless of manifest. Beyond that, the manifest derived from your enabled hubs at provision time scopes what else is on disk. Composition is build-time, not runtime.

proven-core · 15 skills

floor

commit
plan
decide
learn
explain
debug
test
review-pr
refactor
architecture-report
security-audit
health
cognitive
deliberate
multi-track

versionv4.80

load time<200ms

tenant-profile.json · v1

manifest

{
  "version": 1,
  "enabledSkills": [
    "sales-present",
    "c4",
    "code-gen"
  ],
  "enabledExtensions": [
    "odin-compass"
  ],
  "providerProfile": null,
  "_meta": {
    "sourceHubs": ["sales", "coding"]
  }
}

Workflow

/deliberate runs the work. Receipts at every phase.

The flagship multi-round planning loop: 7 phases from preflight to ship report. Every phase emits an AuditEvent. Operator gates between planning and execution; sub-agents bounded by the tenant's manifest, provider, and budget.

terminal · /deliberate phases

+8-0bash

1$ odin-agent /deliberate "add payment retry logic"

3 Phase 0 Preflight → tools detected: codex T5, factory T1, droid T2

4 Phase 0.5 Subject pin → cycle-context written, lock acquired

5 Phase 1 Cognitive → 5-perspective council, 78% confidence

6 Phase 2 WO Planning → 4 WOs drafted, validator GREEN

7 ─ operator approval gate ─

8 Phase 3 Execution → /multi-track team spawned, 4 WOs running

9 Phase 4 DROID sweep → composite 87/100, GREEN to ship

10 Phase 5 Residual → 2 follow-up items filed with disposition

11 Phase 6 Honest report → per-phase bars + composite + cost ledger

13 Total cost: $0.42 · Wall clock: 12m 47s

every phase audited · operator-gated transition Phase 2→3

Bounded

Per-tenant budget. Per-tenant providers. By construction.

The runtime cannot exceed your tenant's daily or per-session budget cap. Provider API keys come from your tenant env, not a shared pool. Spawning a sub-agent counts against your budget, not the operator's.

budget · today

$3.42 / $10

Hard cap. When budget hits zero the runtime refuses new invocations until midnight UTC or operator reset.

remaining$6.58

sessions12

providers · per-tenant

4 active

Anthropiclive
OpenAIlive
OpenRouterlive
MLX (local)live
Codex CLIopt-in

spawn-tree · last cycle

6 spawns

commander · 1 instance
→ feature-architect · T6
→ plan-validator · T3
→ codex-task × 3 · T5
→ droid-reporter · T2

Every action inside your boundary.

Inside the workspace. Inside your budget. Inside your boundary.

Request enterprise access See it for solo + small teams

Sovereign Cloud Workspace

Code in your browser. On your infrastructure. Governed by construction.

Learn more →

Engines

Hub Engines

Six domain-isolated engines that ship code, decisions, content.

Learn more →

Memory

BrainDB

Governed memory. Append-only audit. Semantic recall over the why.

Learn more →

Ops

Control Tower

Operator dashboard for your tenant fleet.

Learn more →

Assistant

Odin Assistant

Voice + chat assistant for the workspace. Multi-provider. Audited.

Learn more →

{ "version": 1, "enabledSkills": [ "sales-present", "c4", "code-gen" ], "enabledExtensions": [ "odin-compass" ], "providerProfile": null, "_meta": { "sourceHubs": ["sales", "coding"] } }