Model reference
Fantail
Fantail is Autohand's latency-first model for focused coding work. It is the default cloud model in Autohand Code.
Availability
Fantail is available now in Autohand Code and through the Autohand API. Use the exact model identifier fantail.
Fantail is included with the Autohand Code Free plan. See plans and pricing for account access details.
Technical specifications
| Property | Value |
|---|---|
| Model ID | fantail |
| Input context | 256K tokens |
| Maximum output | 16K tokens |
| Tool calls | Supported |
| Serving profile | Latency-first coding model |
When to use Fantail
Choose Fantail when the relevant files are already known or easy for the agent to inspect, and the next action benefits from quick feedback.
- Code completions, small implementation tasks, and quick fixes
- Review comments, test scaffolds, and documentation snippets
- Short agent loops where the task stays within a focused area of the repository
Choose Moa when the task depends on broader architecture, many files, or a more deliberate plan.
Use Fantail
Autohand Code and the HTTP API both select Fantail with the model ID. Keep the ID explicit in scripts so the execution path remains repeatable.
Autohand Code
autohand --model fantail --prompt "Find the smallest safe fix for the failing test and apply it." --patch
# In an active session
/model fantail
OpenAI SDK
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.AUTOHAND_API_KEY,
baseURL: "https://api.autohand.ai/v1",
});
const response = await client.chat.completions.create({
model: "fantail",
messages: [
{ role: "user", content: "Find the smallest safe fix for the failing test and apply it." },
],
stream: true,
});
HTTP API
curl https://api.autohand.ai/v1/chat/completions \
-H "Authorization: Bearer $AUTOHAND_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "fantail",
"messages": [{
"role": "user",
"content": "Find the smallest safe fix for the failing test and apply it."
}]
}'Prompting for a short loop
Set a bounded goal, name the relevant test or behavior, and ask for the smallest safe change. A focused prompt gives the agent a clear verification target without loading the task with unnecessary context.
Fix the failing parser test in packages/api. Keep the public response shape unchanged, add coverage for the regression, then report the files and test command used.
Plans and limits
Fantail is included on every Autohand Code plan, starting with Free. It is the only hosted Autohand model on the Free plan. Credits and request limits are set by the plan, not by the model, and are listed in full on the plans and pricing pages.
| Plan | Fantail |
|---|---|
| Free | Included |
| Pro | Included |
| Max | Included |
| Team | Included |
| Enterprise | Included |
See plans and pricing for the full limits.
Capabilities and limits
The published contract covers chat completion requests, model selection, and tool calling. This reference does not claim support for response formats, vision inputs, streaming controls, or sampling parameters. Treat those capabilities as unavailable until documented by the API contract.
Keep AUTOHAND_API_KEY on the server or in your local environment. Do not put API keys in browser code or source control.