Custom AI models and providers in Deskgram 2
Connect your OpenAI-compatible provider to Deskgram 2 AI modules. Set the API address, key and models, select the provider in the module and inspect a small test response.
Custom AI models and providers in Deskgram 2
Connect your OpenAI-compatible provider to Deskgram 2 AI modules. Set the API address, key and models, select the provider in the module and inspect a small test response.
Download Deskgram 2 PlansModule settings
Open Settings → API → Custom AI providers and add a provider. Enter a name, Base URL and API key; add at least one exact model ID and select the default model. The client uses GET /models and POST /chat/completions. Base URL is the API root, such as https://api.example.invalid/v1, not the final chat endpoint. A different protocol needs a compatible gateway. Save and select the provider and model in the AI component. A manual model name selects a provider model; it does not download or train a model. Set the prompt and generation parameters, then inspect the context supplied by the module. io.net, OpenRouter, Mistral and saved custom providers remain available in the interface; DG2 free is currently hidden. External API prices and quotas are separate from the Deskgram subscription. A manual model price of zero may be a default, and zero balance without a supported balance endpoint does not establish free access.
Check before running
When /models returns a nonempty list, the connection test confirms model listing, not generation by the selected model. Test a small generation separately. If no model list is returned and a default model is set, the test makes a short chat/completions request that may consume paid tokens or quota. A manual list alone does not validate the key. Check Base URL, authentication, model ID, permissions and response format; 401/403 indicates an authorization issue to investigate, while 404 requires checking the address and endpoint support. Inspect timeouts and service access separately. Review facts, language, length and links in responses; temperature does not establish accuracy. Listing models, generating text and sending to Telegram are separate outcomes. Mistral and other providers are not promised to be always free; confirm costs and limits with the provider.
Related tasks and guides
Choose a related task or open the detailed guide. Linked Russian guides are marked accordingly.
AI comments for Telegram posts AI conversations in Telegram groups Telegram autoresponder Guide in Russian Support