For enterprises

An AI software factory,
inside your walls.

Agents plan, build, test and ship software in your own AWS account, under your policies and your approvals. Trailboss holds no credentials for it, and nothing reaches production without a person you named.

Trailboss Territory · from $1,500 a month, by contract

Built for teams that answer to security.

The questions a security review asks first have answers in the architecture, not in a settings page.

  • 01
    Runs in your AWS

    A worker in your account does the work and only connects out. Delete the stack and every permission goes with it.

  • 02
    Your model, your terms

    Use Claude on Amazon Bedrock in your account, or your own key, and pay your provider directly. Or prepay Trailboss credits.

  • 03
    SSO and SCIM

    OIDC single sign-on, people provisioned and removed from your directory, roles you control.

  • 04
    An audit trail that holds

    Every agent action, approval and cost, append-only and hash-chained, exportable for your auditors.

Your cloud. Your model.

Start on ours.
Move to yours.

Try Trailboss in minutes on isolated runners we host. When you're ready, install a lightweight worker in your AWS account and the agents run there instead. It only connects out; delete the stack to revoke everything.

Hosted runnersYour AWS
SetupMinutes. No AWS account.A worker stack in your account.
Where agents runAn isolated container per job on our runners.Your AWS: checkout, builds, tests, previews, deploys.
ModelYour own model key.Claude on Amazon Bedrock, or your key.
Production deploysNot from hosted runners.Yes, behind your approvals.
Where your data staysCode runs on our runners for the job.In your account. With Bedrock, prompts and responses too.

With your own API key, prompts go only to that provider, under your agreement with them.

10The boundary
Trailboss cloud
Control plane
  • Dashboard and API
  • Projects, issues, releases
  • Policy engine and approvals
  • Orchestration and audit

Holds no AWS credentials for your account.

signed jobsresults back
Your AWS account
Execution plane
  • Execution worker
  • Isolated agent containers
  • Code, builds and tests
  • Preview environments
  • Secrets and deploys
  • Claude on Amazon BedrockPrompts and responses stay here

Assume every agent can be fooled. Build so it doesn't matter.

Security is the architecture, not a settings page: a policy engine, a credential broker, an execution broker and a lifecycle manager, each separate from the agent.

  1. P-01
    Every agent has an identity

    Short-lived tokens and a per-agent policy, revocable in one click. No agent inherits a human's permissions.

  2. P-02
    The model never decides

    The agent proposes; a deterministic policy engine decides. Prompt injection can't talk its way past it.

  3. P-03
    No keys in our cloud

    Our control plane holds no credentials for your AWS account. Work runs on short-lived, task-scoped roles inside it.

  4. P-04
    Approvals bind to the plan

    Production deploys and IAM changes need a human, tied to the exact plan hash.

  5. P-05
    Production stays read-only

    Fixes arrive as pull requests, never as writes to prod.

  6. P-06
    Cleanup is guaranteed

    TTLs and budgets live outside the agent. A dedicated role deletes; a tag scan verifies.

ledger · every action append-only and hash-chained: which agent, which task, who approved, what it cost.See the audit log
Governance

Agents do the work. People stay in charge.

Every action an agent takes is decided against a policy, recorded, and stopped where your rules say a person decides. The rules are yours to set.

  • A policy for every agent

    What each agent may touch, spend and run is decided on the server before it happens: resources, environments, commands, cost per task. The model proposes; the policy decides.

  • Approvals bound to the plan

    A production change waits for a person, and the approval is tied to the exact change that was reviewed. If the change moves, the approval no longer applies.

  • A second approver

    Critical releases can require two different people: whoever reviewed the security findings can't also approve the rollout.

  • Checks that fail closed

    Secrets, static analysis, dependencies and tests run on every change. A check that didn't run is a failed check, never a pass.

Security review

What your review will ask, answered.

Send us your questionnaire and we answer it in writing. These are ready now, and below them is what we don't have yet, so you hear it from us first.

Where we are, plainly
Compliance reports
We hold no SOC 2 or ISO 27001 report yet, and we don't claim one. We answer security questionnaires in writing and walk your team through the architecture.
Penetration test
An independent test is planned before general availability; it hasn't run yet. Pilot customers get the report and our fixes when it does.
SAML
Single sign-on works over OIDC today (Okta, Entra ID and Google all offer it). SAML is next; tell us if it's a requirement.
Support terms
Response times and uptime commitments are agreed in your contract, not promised on a web page.
Send us your security questionnaire
Pilot

Start with one team, six weeks.

A pilot runs in your AWS account, on work your team already has, with your security team in the room from the first week. You decide with evidence, not a demo.

  1. Week 1

    We connect your AWS account and identity provider with your platform team, and agree the policies agents run under.

  2. Weeks 2 to 5

    One team runs real work through it: a backlog of issues, or one new internal app, from plan to production behind your approvals.

  3. Week 6

    You get the numbers: what shipped, what it cost, what was stopped by a check or an approver, and every action in the audit trail.

  • Runs in your AWS account
  • Your identity provider from day one
  • Remove it by deleting one stack
  • No long contract to start

Bring your hardest
security question.

Tell us about the team and where the work has to run. We'll come back with how a pilot would look for you.