AutoCAD AI assistant
The AutoCAD AI assistant from Bonfire Terminal runs locally on Windows or Mac, using your design context to help manage drawings and projects from messaging apps. Your AI, your machine, keeping engineering files under your control.
What Bonfire can do in AutoCAD
Real actions from the AutoCAD API that a Bonfire agent can run on your behalf, on-device.
| Action | Resource | Method · Endpoint |
|---|---|---|
| List hubs | Hubs | GET /project/v1/hubs |
| List projects | Projects | GET /project/v1/hubs/{hub_id}/projects |
| Get top folders | Projects | GET /project/v1/hubs/{hub_id}/projects/{project_id}/topFolders |
| List folder contents | Data | GET /data/v1/projects/{project_id}/folders/{folder_id}/contents |
| Get item | Data | GET /data/v1/projects/{project_id}/items/{item_id} |
| Start translation job | Model Derivative | POST /modelderivative/v2/designdata/job |
| Get manifest | Model Derivative | GET /modelderivative/v2/designdata/{urn}/manifest |
| Get metadata | Model Derivative | GET /modelderivative/v2/designdata/{urn}/metadata |
AutoCAD agents in action
On your machine, not a SaaS
Bonfire Terminal
- Your keys & data stay on your device
- Perpetual license — no per-seat SaaS fee
- Any connected model — Claude, GPT, Gemini or Grok
- Offline Llama for local, private reasoning
- Driven from chat you already use
Cloud AI middleman
- Keys & data pass through a third-party server
- Recurring subscription, usage caps
- Model locked to the vendor
- Yet another dashboard to log into
AutoCAD — questions
Can Bonfire automate AutoCAD?
Yes. Bonfire runs on your machine, holds your AutoCAD credentials locally, and a connected model (Claude, GPT, Gemini or Grok) drives the AutoCAD API to run the actions you ask for in chat.
Does my AutoCAD data go to the cloud?
Bonfire runs on your desktop and keeps your keys local — it isn't a SaaS holding your data. Only the specific request you approve is sent to your chosen AI model.
Which AI models can I use?
Any connected model — Claude, GPT, Gemini or Grok. A fully offline Llama model runs locally for private reasoning but can't reach live APIs.