AISBF

Authentication and route families

AISBF exposes OpenAI-compatible global routes, direct provider routes, rotation routes, autoselect routes, user-scoped routes, dashboard management routes, and MCP automation routes. API clients normally send a bearer token.

Authorization: Bearer $AISBF_API_TOKEN
Content-Type: application/json

Core OpenAI-compatible endpoints

Minimal chat request

curl -fsS https://aisbf.example.com/api/v1/chat/completions \
  -H "Authorization: Bearer $AISBF_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4.1-mini","messages":[{"role":"user","content":"Say pong"}],"max_tokens":20}'

Direct provider endpoints

Provider config shape

{
  "id": "openai-prod",
  "name": "OpenAI production",
  "endpoint": "https://api.openai.com/v1",
  "type": "openai",
  "api_key_required": true,
  "api_key": "sk-...",
  "default_rate_limit": 0.5,
  "default_max_request_tokens": 24000,
  "default_context_size": 128000,
  "default_error_cooldown": 300,
  "enable_native_caching": false,
  "enable_response_cache": null,
  "models": [{"name":"gpt-4.1-mini","context_size":128000,"max_request_tokens":24000}]
}

Rotation endpoints

Rotation provider entries are weighted. The higher the weight, the more often that provider/model should be selected when available. AISBF tracks failures and can disable failing routes during cooldowns.

{"model":"coding-production","messages":[{"role":"user","content":"Refactor this function"}],"stream":true}

Autoselect endpoints

Autoselect config has model_name, description, selection_model, fallback, available_models, optional capabilities, and privacy/NSFW/semantic classification flags.

User-scoped API endpoints

Use model IDs such as user-provider/openai-prod/gpt-4.1-mini, user-rotation/coding-production, and user autoselect IDs consistently in app configs.

Dashboard and admin route map

These routes are primarily browser/dashboard surfaces, not public app APIs.

MCP endpoints

The source config separates MCP tokens into autoselect_tokens and fullconfig_tokens. Keep full-config tokens out of normal app runtimes.

Request behavior notes

• Streaming: chat requests accept stream:true; provider handlers normalize streaming where supported.
• Chunking: max_request_tokens can force large requests to split into chunks; streaming falls back to non-streaming for chunked processing.
• Context condensation: provider/model/rotation/autoselect defaults can specify context size and condensation behavior.
• Provider-native caching: source config supports Anthropic cache control and Google context-cache configuration fields. Response cache can use memory, Redis, SQLite, or MySQL.
• Adaptive rate limiting: learns from 429 responses using backoff, headroom, jitter, and gradual recovery settings.

Try AISBF

AISBF is open source and also available as a hosted service. During the current testing period, hosted Pro is temporarily available as unlimited access for €2/month.