v2.5.11

AgenticChunking now accepts a custom_prompt parameter, letting you override the default model-driven chunking instructions with domain-specific logic. Rather than relying solely on the built-in heuristics for finding semantic breakpoints, you can now describe exactly how the model should segment your documents — for example, splitting at major section boundaries, preserving clause integrity, or separating structured metadata from body content — making it straightforward to tune retrieval quality for specialized corpora.

‍

Details:

• Pass any string to custom_prompt to override the default chunking behavior; custom prompts are prioritized over built-in instructions
• The default output format constraints are still enforced automatically — custom_prompt only needs to describe the chunking logic itself
• Always pair custom_prompt with max_chunk_size to bound output length; the default max_chunk_size is 5000 characters
• The model parameter accepts any Agno-compatible model, allowing you to route chunking to a smaller or cheaper model independently of your agent

‍

See the Custom Prompts docs for more.