Translation providers
Glossa integrates remote LLM services, Ollama and OpenAI-compatible endpoints. DeepL is available as the initial stage in DeepL Hybrid mode. This table describes implemented roles without ranking model quality by brand.
| Provider | Configuration | Role |
|---|---|---|
| Gemini | API key and model | LLM stages and assessment |
| OpenAI | API key and model | LLM stages and assessment |
| Anthropic | API key and model | LLM stages and assessment |
| DeepSeek | API key and model | LLM stages and assessment |
| DeepL | API key and translation options | Initial translation in DeepL Hybrid |
| Ollama | Server URL and installed model | LLM stages and assessment |
| Custom | Endpoint profile, model and credentials if required | Stages supported by the configured service |
Credentials
Open Settings → Provider. Keys are stored in the operating system’s credential store when available; otherwise Glossa uses an encrypted local store. Keys are not included in application backups.
Requests send the text and context required by a stage to its selected provider. Choosing Ollama for translation does not make other operations local: also check the evaluator, prompt refinement and any memory services. An Ollama server configured on another computer receives requests at that address.
Custom endpoints
A Custom profile contains a name, base URL, authentication requirement and associated key. The name and URL must be valid before saving or testing. Remote endpoints require HTTPS; HTTP is accepted only for localhost, 127.0.0.1 and ::1.
In a stage, select Custom, choose the profile and enter the model identifier. OpenAI compatibility describes the protocol; it does not guarantee support for every option or response format. Use the connection test and a segment trial to verify the configuration.
Ollama
Install Ollama and download a model suitable for your hardware using the instructions for your distribution. Configure its server URL in Glossa and refresh the model list. The server must be running before processing; if your installation does not start it automatically, use ollama serve.
Performance and capabilities depend on the model, available memory and request length. Use reasoning options only with models that support them. A standard local server does not require an API key.
Selection and diagnosis
Evaluate a model on representative passages, considering accuracy, register, glossary compliance, latency and usage. Keep the assessment criteria fixed while comparing configurations. Model availability, quotas and pricing depend on the service; the application’s catalogue does not replace your account’s terms.
See Troubleshooting for connection, quota and response errors.