Agent delivery plane

Trust is the foundation.
Everything the agent does inherits it.

Any coding agent can write code. Steel Mission gives that work a boundary you actually control: no command runs unless policy allows it, nothing executes outside your runtime, and every action leaves signed evidence you can hand to an auditor. Trust isn't a feature added on top — it's the layer every capability stands on.

Install Core Request Enterprise briefing

Mission progressWorking

PRJ-0001 · Durable Core · MS-0002Working

Stage: Test → Inspect
Runner: private · container · attested
Evidence: #0412 signed · chain intact

PRJ-0001 · Release flowWaiting on person

Decision requiredCommand gh pr create is outside baseline policy. Approver: DC13 owner.

Knowledge qualityStale source

KD02 Delivery Doctrine is 14 days past review. Owner notified; claims about intent are withheld until repaired.

How it works

Understand → Close · every stage evidenced

01Understand

Bind the request to organizational knowledge and check freshness, ownership and sufficiency.

02Plan

Delivery Coordinator drafts scope on the bound model; risk tier assigned.

03Modify

The coding agent edits inside an isolated worktree.

04Build

Commands run only through the authenticated private runner.

05Test

Results and artifacts are hashed and attached to the mission.

06Inspect

Static, security, fuzz and platform verifiers review the change.

07Repair

Failures loop back under the same policy envelope.

08PR

Out-of-baseline commands request approval; SoD enforced.

09Deploy

Production paths require external evidence signing.

10Close

Signed proof pack returned to the originating tool.

Originating tools

Customer-controlled runtime

Policy · approval · evidence

Private runnerephemeral hardened container

Trust boundary

Authority never leaves the inner rings.

Requests come in from the tools where work starts. Policy, approval, evidence and execution authority are decided inside your runtime. Only the private runner executes, and only signed results come back out.

Orchestration systems such as n8n may request work or receive events — they are never the source of truth. Evidence is signed locally or by a customer-controlled external signer or KMS.

Editions

Capability	Core · Apache-2.0	Enterprise Edition
Self-managed SSO / OIDC, audit logging, SIEM export	Included	Included
Guarded control plane, private runner, signed evidence	Included	Included
Native GitHub, Slack, Jira adapters · starter company	Included	Included
Multi-organization fleet governance	—	Included
Managed deployment, upgrades, evidence retention	—	Included
Managed KMS/HSM operations · private-cloud operations	—	Included
Advanced separation-of-duties workflows · support	—	Included

Integrations

GitHubSCM adapter · included

SlackChat adapter · included

JiraIssue adapter · included

Claude CodeModel provider · included

CodexModel provider · included

OllamaLocal model · included

GitHub ActionsCI adapter · upcoming

GitLabSCM adapter · upcoming

TemporalOrchestration · upcoming

Compliance

SOC 2Proof packs map mission evidence to trust-service criteria.

ISO 27001Signed, hash-chained records for information-security controls.

ISO 42001AI-management evidence: model binding, approvals, knowledge quality.

FAQ

Is Steel Mission a coding agent?+

No. Claude Code, Codex or a local model writes the code. Steel Mission owns the boundary around that act: whether a command may run at all, where it runs, and what evidence it leaves behind.

Which models can run the Delivery Coordinator?+

Model choice is separate from organizational role. Delivery Coordinator is a capability that can be bound to Claude, OpenAI/Codex, Glimmer or a local model via Ollama (qwen2.5-coder:14b). Binding creates a model instance, not a new role.

Does it depend on n8n?+

No. n8n is a replaceable orchestration adapter that can request work or receive events. It is never the source of truth for policy, approval, evidence or execution authority.

Where does it run?+

Inside your infrastructure or private cloud. Policies, auth, knowledge, evidence and connector configuration are customer-owned, and production execution goes through an authenticated private runner in an ephemeral hardened container.

What is in Core versus Enterprise Edition?+

Core (Apache-2.0) includes SSO/OIDC, audit logging, SIEM export, native connectors, external evidence signing and the starter company — trust controls are not license-gated. Enterprise adds fleet governance, managed deployment, KMS/HSM operations, advanced separation-of-duties workflows, managed retention and support.

Do my teams have to work in a new UI?+

No. Requests originate in repositories, issue trackers, chat, CI, IDEs or provider-native tools, and Steel Mission returns status, approval requests, control decisions and evidence links there. The built-in UI is for configuration, investigation and fallback.

What do I need to try it?+

macOS or Linux, Python 3.11 or newer, Git and a modern browser. Optionally Claude Code CLI, Codex CLI, Ollama, GitHub CLI and Docker for production-eligible private-runner isolation.