Service depth
CLI agents.
Terminal agents for development and infrastructure, by tier, with your models, inside one enforced harness. A developer agent writes and commits. An operator agent approves and deploys. An infrastructure agent plans and provisions. People approve what they can see.
CLI agents
Surface
A terminal that carries the business.
The agent works in its own terminal interface, dressed in the business mark and the business palette.
- Custom TUI and apps.A TUI (terminal user interface) built for the agent, with its own applications for alerts, queues, models and history.
- Your brand in every session.The business mark heads every session and every screen, from the session header to the home screen, on any terminal your teams use.
- Views and palettes.The terminal is themed to the business: twenty layout families and three palettes, black, gray and white, chosen per client, so the terminal and the web application read as one product.
Architecture
The architecture ships with the project.
The full agentic architecture lives in the repository, and the harness enforces it.
- One record.Each agent's persona, catalog, permissions, connections and audit channel are compiled from one core record, and its signature shows in every session header.
- Enforced plugins per category.Interface, audit and path-guard plugins are required by category, with their versions pinned in every session header.
- Workflows and commands.Workflows and slash commands are versioned with the code, so every developer and every agent runs the same steps.
- Enforced project architecture.Dependency rules and the project layout are checked on demand and before changes land. Violations are reported, never waved through.
- Skills and books.Skills carry the project's procedures; books carry the reference knowledge the agent reads before it acts.
Operations
Infrastructure by tier, approved by people.
Tier-level agents do the infrastructure work. Anything consequential becomes an alert that a named person approves.
- Three agents by tier.A developer agent (tier 1) writes and commits, an operator agent (tier 2) approves and deploys, an infrastructure agent (tier 2) plans and provisions. No agent acts outside its tier.
- Alert approval.Approvals wait in the terminal's alerts list with the change, the agent and the target. One keystroke approves.
- A2A (agent to agent) protocol.Tiers hand tasks to each other over A2A, and every task reports its state from submitted to completed.
- Task queue management.Queued, running and finished tasks are listed per project, each with its owner and state.
- Audit channels.Every tool call, refusal and approval lands in the agent's audit channel: a tool log for the developer tier, a hash chain for the operator and infrastructure tiers, verifiable on demand.
- Workers and sessions.Start, stop, resume and audit every worker and session.
Models and runtime
Logic first, models after.
Deterministic rules decide what they can. Models are managed, tested on the project, and run where you choose.
- Managed models, ready per project.Models are tested on the project's own tasks and marked ready before an agent may call them.
- Logic first, AI after.Every task passes deterministic prechecks before a model is called. Where a rule can decide, the rule decides.
- Local, hybrid or API token.Run models on a local endpoint, mix local and remote, or connect through an API token.
Where to go next
Watch the three agents in live scenarios, or read how the whole service is delivered.