Repository navigation
AI Providers
AI is optional and off by default. Settings → AI analysis (Türkçe: AI analizi).
| Provider | Group | Key needed | Notes |
|---|---|---|---|
| Ollama | Local | no |
http://localhost:11434/v1 — free, runs on your PC |
| LM Studio | Local | no | http://localhost:1234/v1 |
| Google Gemini | Free tier | yes | |
| Groq | Free tier | yes | |
| OpenRouter | Free tier | yes | |
| NVIDIA NIM | Free tier | yes | pick a model with Fetch models |
| Mistral | Free tier | yes | |
| Anthropic Claude | Paid | yes | Messages API |
| OpenAI | Paid | yes | uses max_completion_tokens
|
| Kimi (Moonshot) | Paid | yes | |
| xAI Grok | Paid | yes | |
| EVREN LLM | Other | yes |
X-API-Key header |
| Alibaba Qwen (Model Studio) | Other | yes | enter your workspace URL |
| OpenAI-compatible (custom URL) | Other | yes | any server speaking /chat/completions
|
- Choose the Provider; enter the key if needed (get a key opens the provider page).
- Fetch models — lists the models the provider really offers (no tokens spent). Pick one or type its name.
- Test connection — one tiny real request (~30 tokens).
- Save.
Model ids age quickly, so Draxbyte does not hard-code them: it asks the provider (GET {base}/models). A few providers have a starting default.
Off by default · per-request approval showing the exact text · local cache (path + size + modified time) · cache is mandatory · daily token limit (clearing the cache does not reset it) · tiny requests (max_tokens 200, no streaming).
Model listing and chat were tested against a real local OpenAI-compatible server. Other providers are covered by unit tests (request shape, auth header, response parsing) but not tried with live accounts.
https is required except for local servers (Ollama, LM Studio, loopback addresses). Keys are redacted from error messages. They are stored as plain text in config.toml; prefer the DRAXBYTE_API_KEY environment variable.
English
Türkçe
Links