Two specialists.
One execution loop.
Altr doesn't use general-purpose chatbots. We use specialist agents designed for high-precision product work: one to structure the intent, and one to build the change.
Structures the work.
The Spec Agent is responsible for context preservation. It reads Slack threads, identifies feature requests, drafts acceptance criteria, and produces a spec that stays attached to the code forever.
Capabilities
- Captures Slack threads verbatim
- Drafts specs with acceptance criteria
- Flags unresolved decisions before build starts
- Writes the spec live — streams into the editor
- Co-authors release notes after merge
Ships the change.
The Build Agent handles the execution. It opens isolated git worktrees, proposes implementation steps, runs tests, and opens PRs with the original spec still attached.
Capabilities
- Opens isolated git worktrees per ticket
- Proposes implementation steps before code
- Streams progress back to the ticket view
- Opens PRs with spec and criteria attached
- Checks implementation against original intent
Interactive
Autonomy per-ticket,
not per-agent.
You decide how much control to delegate for every task. Whether you want a specialist to draft while you code, or ship while you review, the choice is made at the ticket level.
The Golden Rule
“Human review stays the default gate. Every PR opened by an agent waits for your explicit approval before it merges.”
Copilot mode
Agent proposes implementation steps. You review and pair on the diff.