Docs
First session
After sign-in you land in chat. Previous projects on this account load automatically.
- Connect Rhino or SketchUp from the Connect pill. Open the host if it is not already running.
- In Settings (
/model), pick a provider and paste your API key. Keys stay in Windows Credential Manager. How to mint one: API keys. - Set the plot: select a closed curve and
/selectboundary, or tell Urbanaut which layer is the boundary. - Describe a massing, or attach a reference image. Enter sends. Shift+Enter is a new line.
- Read the preview. Apply builds in the host. Discard leaves the model unchanged.
Bring-your-own-key
Urbanaut does not sell tokens. You pay Google, OpenAI or Anthropic (or run Ollama / LM Studio locally). Paste the model id in Settings. Image ids draw pictures; they do not build geometry. Pick a text/tool model for massing.
Suggested agents
For a first session, use gemini-3.8-flash or claude-sonnet-5 at Medium usage. Both follow tool calls well enough to loft, extrude and stay inside the plot. Flash is the cheap daily driver. Sonnet is the safer CAD workhorse when the brief has constraints. Do not start on GPT-6 Astra or Claude Fable. They are five to ten times the cost of Flash for the same Apply/Discard loop.
Estimates below are USD list prices (September 2026) for one typical massing session: plot set, one scheme, two viewport checks, Apply, one revision, Medium usage (~100k input / ~20k output including thinking). Light is about half. High or Studio, long chats, and attached PDFs multiply it. Gemini 3.8 Flash intro rates hold through 31 Dec 2026, then double.
| Model | Best use in Urbanaut | Est. USD / session |
|---|---|---|
| Start here Gemini 3.8 Flash gemini-3.8-flash | First massing and daily CAD agent. Fast tool loops, cheap enough to iterate. Intro price until end of 2026. | ~$0.15 |
| Start here Claude Sonnet 5 claude-sonnet-5 | Urbanaut’s Claude default. Best everyday CAD: plot constraints, ordered builds, fewer invented numbers. | ~$0.40 |
| Daily GPT-5.6 Terra gpt-5.6-terra | OpenAI workhorse. Same job as Sonnet if your key is already on OpenAI. | ~$0.45 |
| Daily Gemini 2.5 Pro gemini-2.5-pro | Still the in-app Gemini default. Solid spatial composition if 3.8 Flash is not on your key yet. | ~$0.35 |
| Daily Gemini 2.5 Flash gemini-2.5-flash | Cheaper Gemini if you already have it. Fine for simple lofts; step up when the scheme gets layered. | ~$0.08 |
| Daily Gemini 3.5 Flash gemini-3.5-flash | Stronger Flash for longer agent sessions. Priced closer to Pro. Use when 3.8 is not listed. | ~$0.35 |
| Heavy Gemini 3.1 Pro gemini-3.1-pro | Long briefs, mixed PDFs + viewport. When Flash drops a setback or mixes layers. | ~$0.45 |
| Heavy GPT-5.6 Sol gpt-5.6-sol | Harder massing: cores, section, several options in one chat. Not for rename-a-layer. | ~$0.80 |
| Heavy Claude Opus 5 claude-opus-5 | Developed design on High/Studio usage. Follows “do not invent FSI” more strictly on long loops. | ~$1.00 |
| Rare GPT-6 Astra gpt-6-astra | Competition-level session with a dense brief. Same Apply/Discard loop as Flash. Pay this only when the reasoning is the work. | ~$2.00 |
| Rare Claude Fable 5.1 claude-fable-5.1 | Same tier as Astra. Long-horizon agent with cache-heavy chats. Overkill for a first massing. | ~$2.00 |
| Skip GPT-5.6 Luna gpt-5.6-luna | Too small as the CAD brain. Use later for /editlayer, takeoff summaries, renaming. Not for lofting a scheme. | ~$0.04 |
| Skip Claude Haiku 4.5 claude-haiku-4-5 | Fast classification, not spatial composition. Drops tools on Medium+ massing. | ~$0.20 |
| Skip Local (Ollama / LM Studio) llama / qwen (local) | Zero token bill. Useful offline for chat and layers. Weak for first-session massing tool loops. | $0 |
Est. USD / session is what you pay your AI provider for one typical massing, not the Urbanaut plan. USD list rates, September 2026. Light is about half; High or Studio, long chats and PDFs cost more.
Concept pictures (/render) are a separate image model, not the agent. Skip any id that contains -image. Hosted Studio renders use the UrbanEyes Gemini image key and count against the Studio quota. Local Ollama / LM Studio is $0 after the machine, but tool loops are weaker. Fine for layer edits, not a first massing.