Text conversation and parameters
Select a text model, configure system instructions, sampling, output and reasoning, and handle streaming and conversation history.
Updated
Before you start
- Sign in to your own LINK42 account. The console uses a login session; programmatic calls use a LINK42 API key.
- Generation is billable. Check balance, model availability and the estimate first. Screenshots only illustrate the steps; they do not show that a model is enabled or what it will generate.
Start a conversation
Select a text model
Choose Text in Experience Center and open the selector. Use an available text model, not an image/video generator for chat/completions.
Provide a task and context
Describe the task. Multi-turn history contributes input tokens; start a new conversation for a different task rather than only clearing the composer.
Review and send
Inspect temperature, output limit and reasoning in the settings panel. The estimate is not the final charge.
Recognize completion and failure
Streaming arrives incrementally. An unfinished stream is not a complete answer. Keep the request ID and reconcile interrupted requests before resending.
View full-resolution imageUnderstand parameters
View full-resolution image| Parameter | Meaning | Constraint |
|---|---|---|
| temperature / top_p | Sampling controls, not an accuracy switch | Use the model's allowed range, usually adjusting one at a time. |
| max_tokens / max_output_tokens | Output token budget, not input length | Bounded by output/context limits; too small can truncate the answer. |
| system | Role, format and behavior instructions | Be explicit, distinguish it from user input and omit secrets. |
| thinking / reasoning_effort | Reasoning behavior for supporting models | Not universal; reasoning tokens may contribute to usage. |
| tools / response_format | Tools or structured output | Requires model capability, protocol support and valid schema. |
Vision understanding is not generation
Only vision-capable text models accept image questions. Originals may be up to 100 MiB, but browser processing and model limits still govern format, pixels, count and request size. A valid text-only key does not grant vision capability.
Troubleshooting
| Symptom | What to check | Next action |
|---|---|---|
| Delayed answer | Queue, reasoning and upstream state | Observe actual status; retain the session rather than guessing progress. |
| Truncated answer | Output limit, finish_reason and context | Increase a supported budget or split the task; do not label it complete. |
| Unsupported parameter | Model and protocol capabilities | Remove the field or use an eligible model; key validity is not capability. |
| Unexpected cost | History, reasoning and output | Reconcile tokens, cache and charge by request ID. |