`orqvion agent`
Pass a natural-language task. The agent reads the repo, plans steps, and proposes edits — stdout-friendly for logs and CI artifacts.
CLI
The same agent engine — headless in your terminal.
`orqvion agent` runs the desktop ReAct orchestrator on the machine where you invoke it — same ONLINE · OFFLINE · AUTO routing as the IDE (online entitlement/BYOK or offline ORQVION AI + GGUF). No Ollama app. No ORQVION-hosted runner pool.
Command
One binary, same permissions model.
The CLI is not a separate cloud product. It respects the same write-approval, sandbox, and org policy settings as the GUI.
Same routing as the IDE
Routing mode, model catalog, and entitlements match Settings in the desktop app. Offline uses llama.cpp on downloaded GGUF — not `ollama serve`.
Workspace-bound
Runs against the directory you point it at. Prompts and source stay local unless your routing mode calls a provider you configured (BYOK).
Use cases
Where teams reach for the CLI.
Not a replacement for the IDE — a complement when the terminal is already home.
- Nightly local scripts that lint, fix, and leave a diff for human review in the morning.
- Pre-push loops on a developer machine — same agent command, reproducible in scripts.
- SSH on a build box where GUI forwarding is unavailable (online or offline routing per policy).
- Batch refactors across monorepo packages with checkpoints and structured logs.
Honest limits
What the CLI does not pretend to be.
Headless agent on your hardware — not a remote ORQVION worker farm or a silent repo upload pipe.
No ORQVION cloud agents
Tasks execute on hardware you control. There is no ORQVION-hosted remote worker pool in this release.
Approvals still matter
Non-interactive modes respect policy. Destructive writes can still require approval or mandatory Docker sandbox.
Same entitlements
Full agent loops require Pro+ on the license plan. Free tier aligns with read-only Ask-style CLI use — see Pricing.