Skip to content

Add Coot Assistant - #395

Draft
Dialpuri wants to merge 12 commits into
pemsley:mainfrom
Dialpuri:add-coot-assistant
Draft

Add Coot Assistant #395
Dialpuri wants to merge 12 commits into
pemsley:mainfrom
Dialpuri:add-coot-assistant

Conversation

@Dialpuri

@Dialpuri Dialpuri commented Aug 18, 2026

Copy link
Copy Markdown
Contributor

Description

This pull requests add the infrastructure for a Coot Assistant which can aid interaction with Coot using a local LLM.

This work requires two models available via Ollama (~10 Gb)

  • embeddinggemma
  • gemma4 with 32k context

This can be made with this command :

echo -e "FROM gemma4:12b\nPARAMETER num_ctx 32768\nPARAMETER temperature 0.7" > Modelfile && ollama create gemma4-long -f ./Modelfile && rm Modelfile

Once they are pulled it should detect them in the drop down for use in the Coot assistant interface.

Also included in this PR:

  • Fix for RPC spawning which didn't allow multiple connections
  • Added markdown to VTE
  • Added autozoom function to Coot commands

Dialpuri added 12 commits July 24, 2026 10:31
…istant output

- Updated test cases to include `playbook_k` parameter in agent calls for consistency.
- Added new tests to handle empty responses from the agent, ensuring retries and fallback mechanisms work correctly.
- Implemented Markdown rendering in the assistant output, allowing for formatted text (bold, code, headings) to be displayed correctly in the transcript.
- Introduced new tags for Markdown elements and ensured they are applied during text insertion.
- Enhanced the assistant's event handling to include playbook notifications and warnings in the output.
- Improved UI layout for the assistant tab, ensuring proper spacing and organization of input and control elements.
- Updated the default model to "gemma4-long" and modified the system prompt to improve interaction.
- Implemented streaming support in the _ollama_chat function to allow real-time reasoning updates.
- Added new functions to merge tool call deltas and assemble streamed responses for better handling of OpenAI server-sent events.
- Enhanced the run_agent function to manage streamed reasoning and emit thinking events appropriately.
- Introduced a new focus module to manage view adjustments automatically based on the current residue or selection.
- Updated navigation commands to utilize the new focus functionality for better user experience.
- Added clash handling in validation commands, allowing users to navigate to specific clashes found during validation.
- Improved playbook instructions for handling clashes, emphasizing step-by-step execution and reasoning.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant