HTQL CLI

Your AI agents. Under your rules.

Run Claude Code and Codex on your own computer, give each agent a role your business defines, and keep a record of every piece of work.

  • Your own AI provider
  • Roles you define
  • Every agent’s work on record

The problem

AI can write the code. Nothing says who it works for.

When a team runs several agents, who is doing what, and what happens to the work when a session ends?

No role

A prompt is not an authorised job.

No record

The reasons behind the work end with the session.

No control

Stopping an agent means starting again.

Roles first

Every agent starts with a role. Then it gets the work.

Work is given to an approved role, with an ordered choice of AI providers. Each agent accepts work in its own working copy, reports progress and can be paused and resumed.

The agent knows its job before it sees the code.

The roles

Fourteen roles, each with a job description.

Mission, authority, boundaries and completion criteria, approved before any AI works in the role.

Plan and decide

  • Project Manager
  • Product Manager
  • Board of Directors

Design

  • Solution Architect
  • Lead Architect
  • Security Engineering Specialist

Build and verify

  • Software Engineering Agent
  • Engineering Lead
  • QA Engineer

Ship

  • Technical Writer
  • Product Assurance
  • Merge Engineer
  • Release Engineer

Customers

  • CustomerCare Specialist

Reception brings work in and is the customer contact point. It is not an engineering role.

How work moves

From an issue to a release, on the Blackboard.

Your team can see every piece of work and where it stands. Nothing skips a step.

AI recommends and builds. People approve features, merges and deployments.

Approved workflow design

  1. 1. In

    Reception and project planning.

  2. 2. Decide

    Assess customer-visible changes.

    Human approval
  3. 3. Design

    Specify and write protected acceptance tests.

  4. 4. Build

    Implement test first, in its own working copy.

  5. 5. Verify

    Independent code review and QA.

  6. 6. Prepare

    Documentation, architecture and claims.

  7. 7. Merge

    Check approvals and prepare the merge.

    Human merge
  8. 8. Release

    Publish the release.

    Human deployment approval
  9. 9. Close

    Report and notify.

What it does

Everything a team needs to trust its agents.

Bring your own AI

Claude Code or Codex; provider fallback.

Roles your business defines

Activate, disable or add roles.

Each piece of work on its own

Separate repository working copies.

Reports, not chat logs

Checkpoints, questions and results.

Pause and resume

Pick up where work left off.

Works offline

Sync the Blackboard when connected.

Your repository

GitHub or GitLab.

One sign in, every business

Switch between permitted businesses.

Unattended machines

Owner-managed app credentials.

Usage as reported

Actual provider usage, never estimated.

How it works

Four steps to your first agent.

  1. 01Sign inCreate your account, take a plan, and download the CLI.
  2. 02Connect your repositoryPoint the CLI at your GitHub or GitLab repository.
  3. 03Choose your rolesSwitch on the engineering roles your team uses, or add your own.
  4. 04Give it workAgents take the work, report as they go, and leave a record your team can read.

Plans

Free for one developer. US$19 for each on a team.

You pay your AI provider directly. Sigmar never charges for AI use.

Compare plans →Save with Sigmar One →

Team

US$19 / developer / month

Shared roles, synced Blackboard, agent work record.

Buy Team

Put your agents to work. Under your rules.