# Documentaziun per agents d'IA

Questa documentaziun è concepida per vegnir legida tant dad assistents d'IA sco da persunas. Dà a tes agent in dals puncts d'entrada sutvart e
el po scriver code d'integraziun correct senza stimar nums da parameters u pretschs.

## Puncts d'entrada

| URL | Cosa ch'igl è |
|---|---|
| [https://doitong.com/developers/llms.txt](https://doitong.com/developers/llms.txt) | Index en il format [llms.txt](https://llmstxt.org): mintga guida e model cun ina descripziun d'ina lingia. |
| [https://doitong.com/developers/llms-full.txt](https://doitong.com/developers/llms-full.txt) | Mintga guida cumpleta, ils models relevants cumplets e las infurmaziuns essenzialas da tut ils 1 496 models (id, producent, pretsch, link a la pagina cumpleta) en ina datoteca Markdown. |
| `/<lang>/developers/llms.txt` | Il medem index cun las guidas en in'autra lingua dal site, per exempel `https://doitong.com/de/developers/llms.txt` (englais è la lingua standard). |
| `<any page>.md` | Il dubel Markdown d'ina pagina, per exempel [https://doitong.com/developers/models/flux/schnell.md](https://doitong.com/developers/models/flux/schnell.md) u [https://doitong.com/developers/docs/quickstart.md](https://doitong.com/developers/docs/quickstart.md). |
| `/developers/models/<id>/llms.txt` | Il medem sco il dubel `.md` dal model. |
| [https://doitong.com/developers/openapi.json](https://doitong.com/developers/openapi.json) | OpenAPI 3.1 per generaturs da code: mintga endpoint e corps da dumonda tipisads dals models relevants; `?model=<id>`, `?category=<id>` u `?models=all` agiuntan auters. |
| `https://api.doitong.com/v1/models` | Il catalog sco JSON (naginas clavs necessarias). |

La pagina Markdown da mintga model ha ses pretsch en USD e credits API, mintga parameter cun tut las valurs permessas, in exempel mo cun ils parameters necessaris ed in
exempel cumplet, la furma da la resposta e code pront per curl, JavaScript, Python e MCP.

## Copiar per IA

Mintga pagina da model ha in buttun **Copiar per IA**: copia la pagina sco Markdown u avra ella en ChatGPT u Claude cun in
prompt che mussa l'assistent a la versiun Markdown.

## In prompt che funcziuna

Engolescha quai en Claude, ChatGPT u Cursor, e descriva lura tge che ti vuls construir:

```text Prompt
I want to generate images/video/audio with the Doitong API.
Read https://doitong.com/developers/llms.txt first, then the .md page of each model you plan to use.
Facts: base URL https://api.doitong.com/v1; auth header "Authorization: Bearer <key>" with the key from the
DOITONG_API_KEY environment variable; run with POST /run/{model_id}?wait=20 and, if the status is
still queued or running, poll GET /requests/{id}?wait=20; outputs[].url are file URLs on Doitong storage
(kept forever by default; add ?storage=temporary for throwaway runs, deleted after 7 days).
Get the price first with POST /quote/{model_id} (same body): exact, or for a model charged by usage the
most the run can cost (the unused part is refunded). Prices are in USD; 1000 API credits = 1 USD.
Use the parameter names from the model page; never invent parameters.
```

## Laschar l'agent clamar models sez

Per laschar in agent generar medias directamain, connecta il [server MCP](https://doitong.com/developers/docs/mcp) cun ina clav che ha in
budget en credits API. L'agent ha lura utensils per tschertgar models, citar pretschs, exequir e retrair resultats, e na po betg spendrar dapli che
il budget.