explore(ai): tool-using topology generation (shelved; reopen if the deterministic fixes plateau)

Open
#830 0 comments 0 reactions 1 assignee View on GitHub

Maintainers usually reply within 1 day

@vendrabuck is already working on this.

Since Sep 15, 2026.

Assessment

This issue has not been assessed yet.

Description

area:ai exploring

Part of the AI builder wireability epic (#825). Diagnosis option 5, deliberately SHELVED: filed so the reasoning is not lost, not for scheduling.

Idea

Convert generate_topology from one forced propose_topology call (services/ai-orchestrator/app/services/ai_client.py:538-541) into a tool-using loop where the model can call get_device_ports and find_path (both exist in app/services/tools.py for the reservation assistant) before proposing, reusing the assistant loop machinery (answer_reservation_question_with_tools, assistant_max_tool_iterations default 8).

Why shelved

  • The local vLLM model used for evaluation is slow (about 26 s median per call in the June assistant rehearsal) and weaker models handle multi-step tool loops poorly; a tool loop multiplies latency by the iteration count.
  • The deterministic path (cabling-aware resolver plus commit-time validation plus prompt grounding) gets the same outcome without depending on model behavior.

Reopen when

Phases 1 and 2 plateau on the harness, or a hosted model becomes the default provider and latency stops mattering.

Dominant language
Python
Stars
3
Forks
2
Avg merge
20h 19m
Merged PRs (30d)
103

Getting set up

  • Ships a Dockerfile or Docker Compose file
  • No pull request template
  • No contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from vendrabuck/HERD

All issues in vendrabuck/HERD

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.