Refactor the agent service around native messages, skills, MCP tools, a UI message stream and a shared session

Abierto
#9,453 1 comentario 0 reacciones 1 asignado Ver en GitHub

Los mantenedores suelen responder en 1 día

@bobbai00 ya está trabajando en esto.

Desde el 15/11/2026.

  • #9494 de @bobbai00 — abierto
  • #9495 de @bobbai00 — abierto
  • #9496 de @bobbai00 — abierto

Evaluación

Este issue todavía no se ha evaluado.

Descripción

Task Summary
Target design of the agent

This refactor reshapes the agent service around four ideas.

  1. Standard tool-call state. The agent keeps tool calls and their results in the AI SDK's own message history, instead of rebuilding a summary by hand at every step (#9455).
  2. A few tools, with edits as patches. The agent reads a workflow, looks up operators, executes, and edits with one tool whose input is a patch: one call can add, modify and delete several operators and their links, and is applied whole or not at all (#9481). Later the same tools are served from a Texera MCP server, one of many the agent can connect to (#9457).
  3. Skills. The system prompt shrinks to the essentials, and Texera's know-how moves into skills loaded only when needed (#9456).
  4. One protocol to the browser, and one shared session. The agent streams its turns to the browser in the AI SDK's own UI message stream, which the SDK's chat clients read in Angular today and in React later (#9482). Instead of editing its own copy, the agent joins the workflow's shared editing and execution session, so the agent and the user see the same workflow and the same executions (#9464).

The order: #9454 first, so every later step can report its score. Then the loop, the tools and the protocol: #9455, then #9481 and #9482, then #9456. Then the shared session (#9464), and last the MCP server (#9457), whose tools edit the shared workflow.

Saving conversations with an owner (#5267, #5268) moves to the resource unification in #9395, where a conversation becomes a resource kind. In the picture, the session store comes with that work, and add_operator and edit_operator are one edit_workflow tool.

For background, see #9377 and #9395. A few open fixes should land first: #9149, #9259, #8753 and #9043.

Task Type
  • Refactor / Cleanup
  • DevOps / Deployment / CI
  • Testing / QA
  • Documentation
  • Performance
  • Other
Lenguaje dominante
Scala
Estrellas
333
Forks
202
Merge medio
3 d 11 h
PR fusionados (30 d)
185

Preparar el entorno

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de apache/texera

Todos los issues de apache/texera

Issues similares

Más issues de Scala

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.