AigentFlow Desktop is in pre-release. This documentation reflects the current build and will change before general availability.
Reference
Models and effort
Which model runs the work, and how hard it thinks.
AigentFlow works with two providers, Claude and Codex. You choose the model per session, and separately per team member, so a reviewer can run on a different model from the engineer whose work it is reading.
Claude
| Option | Description |
|---|---|
| Fable latest | Deepest reasoning |
| Opus latest | Most capable. The default |
| Sonnet latest | Balanced |
| Haiku latest | Fastest |
| Opus 4.8 | Pinned to that specific version |
The four "latest" options follow whatever the underlying tool currently offers, so they move as the models move. The pinned option does not.
Codex
| Option | Description |
|---|---|
| GPT-6 Astra | Newest frontier model, for the hardest work. The default |
| GPT-5.6 Sol | Frontier model for deep coding and reasoning |
| GPT-5.6 Terra | Balanced, for everyday work |
| GPT-5.6 Luna | Fast, for lighter tasks |
Effort
Four levels: Low, Medium, High and Max. High is the default and matches the provider’s own default depth. Low is fastest and cheapest, Max is slowest and most expensive.
Presets
Rather than choosing per activity, pick a preset and let it decide.
- Maximum — the most capable model on each provider, at maximum effort, for everything.
- Balanced — top models where judgement matters, mid-tier for routine work.
- Economy — fastest and cheapest, for light edits, quick lookups and high volume.
Cost
The app prices each turn from the tokens it used and the model that ran it, and shows the running total for the session. Type /cost at any time.
What your AI provider charges is their bill, not ours. AigentFlow shows you what a session cost so you can see it, and does not collect it.