DOCS
Base URL https://api.verdantbloom.bar/v1. Any OpenAI-compatible client works.
QUICKSTART
- Create an account and make a key on your account page. The key is shown once.
- Set it in your environment:
export VB_KEY=vb-live-... - Call the API:
curl https://api.verdantbloom.bar/v1/chat/completions \
-H "Authorization: Bearer $VB_KEY" -H "Content-Type: application/json" \
-d '{"model": "mistral-nemo-12b", "messages": [{"role": "user", "content": "Summarise: ..."}]}'
import os
from openai import OpenAI
client = OpenAI(base_url="https://api.verdantbloom.bar/v1", api_key=os.environ["VB_KEY"])
reply = client.chat.completions.create(
model="mistral-nemo-12b",
messages=[{"role": "user", "content": "Summarise: ..."}],
)
print(reply.choices[0].message.content)
OPENAI COMPATIBILITY
POST /v1/chat/completions: the OpenAI chat body.stream: truestreams server-sent events ending indata: [DONE]. Vision models takeimage_urlcontent parts.max_tokensabove your level's limit is lowered to it.GET /v1/models: the models your key may call, withcontext_length.- Reasoning: models that can think have it off by default; turn it on with
"chat_template_kwargs": {"enable_thinking": true}. - Not offered: embeddings, image generation, audio, fine-tuning, files and batches, the legacy completions route.
- A request that has to start a model and is streamed gets its 200 at once, then SSE comment lines (
: vb-progress ...) every 5 s until the first token; OpenAI clients ignore them.
TIME TO FIRST TOKEN
GET https://api.verdantbloom.bar/v1/taps/<model>/eta (with your key) answers {"model", "state": "warm|booting|cold|full", "eta_s", "eta_max_s", "queue_ahead", "progress"}. It never starts a model and never counts toward limits. Models stay running 10 minutes after the last request.
ERRORS
| STATUS | CODE | MEANING |
|---|---|---|
| 401 | no_key / bad_key / revoked_key | missing, unknown or revoked key |
| 402 | tab_dry | your monthly credit cannot cover the request |
| 403 | model_not_in_level | the model is not in your level |
| 400 | context_ceiling | prompt plus max_tokens exceeds your level's context |
| 409 | changing_the_keg | Free level: the model is not running right now (not counted) |
| 429 | rpm / parallel / daily_cap | a level limit; retry later |
| 503 | no_capacity | no GPU free for that model right now; nothing charged |