Tools, Memory, And Outputs#

Agent Master is built to be auditable and reproducible: every run leaves a trace, figures are collected in one place, generated code is validated before it is shown, and sessions can be saved and restored. This page covers those outputs.

Step Traces And Provenance#

Each completed request carries a steps trace — an expandable record of the steps the workflow ran, with timing and the LLM cost of the request (often $0.000000 for local processing). The response also states the routing decision and the outcome, for example Workflow ‘orchestrated_report’ complete: 3/3 steps succeeded in 3.3 s. Together these make a run reproducible and reviewable rather than a black box — you can see exactly what was done.

The Figures Panel#

Figures produced by any workflow are collected as thumbnails in the Figures panel of the History sidebar, and each response notes how many were produced (5 figures generated — open the Figures panel to view). The panel accumulates across the session, so a survey’s pseudosections, strike roses, phase-tensor plots, and QC images build up in one place as you work. (The sidebar Figures panel is visible in the workflow screenshots in Workflows And Agents.)

Figures are saved in the format set by Default figure export in Settings — for example PNG (300 dpi).

Generated Code#

Ask Agent Master to generate the code for a workflow and it returns a standalone, runnable pyCSAMT script — the reproducible counterpart of the interactive run.

Agent Master returning a validated standalone pyCSAMT script in a code block

A code-generation request returns a standalone script that reproduces the workflow, with a Copy button and a validation note — Validated: syntax OK and all pyCSAMT imports resolve to real symbols.#

Two things make this trustworthy:

  • the script is validated before it is shown — the syntax is checked and every pyCSAMT import is confirmed to resolve to a real symbol, so it is not hallucinated API;

  • it is self-contained — a header records that it was generated by the pyCSAMT CodeGenerationAgent and how to run it, and you edit only the data path.

Copy it out and run it to reproduce the workflow headlessly or fold it into a larger script.

Sessions And Pinned Prompts#

The History sidebar keeps your work:

  • Chat / Session tabs switch between the current message history and saved sessions; Save on the command bar persists the session.

  • New Chat starts a fresh conversation without unloading the survey.

  • Pinned prompts are kept for reuse across chats — handy for a request you run often (for example plot the pseudosection phase).

Because loaded data, figures, and traces live in the session, saving and restoring gives you the whole context back, not just the text.

What To Keep#

For a reproducible record of a session, keep together:

  • the original EDI survey folder;

  • the saved session (conversation, traces, and figure references);

  • any generated scripts you exported;

  • the figures you exported from the Figures panel.

Next Steps#