Fetched documentation exeeds context window limit - Automatic truncation possible? #1212

When opencode pulls documentation from websites, the resulting response can sometimes exceed the context length of the current model in use (currently Claude sonnet 4 for me). It's impossible to continue this sessions in this case. Manually editing the sessions logs and removing the content makes it possible to resume the session.

Could we implement a token counter that considers how many tokens are left in a given session and truncates the fetched content to a size that still fits the context window (including a short hint for the model)?

Here is the error message for reference:

AI_APICallError: input length and max_tokens exceed context limit: 184714 + 32000 > 200000, decrease input length or max_tokens and try again