Native AI across the workbench — not a bolt-on chat tab.
Three modes for three kinds of work — pick the one that fits your intent.
Get answers without touching the repo.
Ask reads your codebase to answer questions, explain code, and surface concepts — purely read-only. No edits, no commands, no side-effects.
Try it for
Think before you build.
Plan creates a structured implementation plan — step-by-step, with proposed edits shown for review. Nothing is applied until you say so. Use it to design before you commit to code.
Try it for
Full execution with guardrails.
Agent reads, writes, searches, and runs terminal commands — everything needed to complete a task end-to-end. Every file change lands in a review panel before it touches your repo.
Try it for
Pick speed vs depth — Altus routes to the right model behind the scenes. No provider keys in the IDE.
Snappy replies for search, explainers, and small edits. Lowest credit burn.
Best for: Quick questions, inline help, light refactors
Balanced quality for everyday chat and Agent turns across your project.
Best for: Daily coding, multi-file guidance
Strong reasoning for complex Agent workflows, planning, and debugging.
Best for: Feature work, architecture, hard bugs
Maximum depth for the hardest problems — premium credit burn, best results.
Best for: Large refactors, subtle bugs, deep reviews
Every workspace root, not just the active editor tab.
Agent resolves paths across all folders in your workspace.
Attach terminal output and files so the model sees real state.
See what changed before you commit to it.
Per-file diffs with Keep all, Undo all, or Review in the editor.
Terminal runs from the agent — dismissible when you are done.
Stay in flow when conversations grow.
Summarises older messages so context stays within model limits — send is rarely blocked.
Status bar shows estimated usage; manual /compact when you want control.
Sensible scaffolds for Next.js, Flutter, and more — injected into Agent prompts.
Shipped JSON, optional local override, or team-hosted playbook URL.
