PyHydroGeophysX.agents.assistants.geosage package#
Submodules#
PyHydroGeophysX.agents.assistants.geosage.tools module#
GeoSAGE’s tools: one per stage of the workflow, with what each needs and makes.
Each tool is offered to the controller only once what it requires exists,
so the order below is a consequence of the data, not a script. The handlers
are placeholders for the port from ZhengyangFang/GeoSAGE;
each says what it has to do and what it must return. A handler takes the
RunContext and returns
(summary, outputs): one sentence a user can read, and a dict of the
artifacts named in produces. ctx.config holds the configuration from
workflow.configure(); ctx.get(key) reads an earlier tool’s artifact;
ctx.output_dir is where files go; ctx.settings has the model access for
the LLM-based agents (api_key, model, llm_provider).
The artifact keys are the contract between the stages:
field_datagridded gravity and magnetic data on the study area,with bounds, fields and uncertainties (DataAgent)
geological_priorslithologic classes, physical-property ranges andgrouping rules, or None when none are available (PetrologyAgent)
property_modelsdensity and susceptibility on the inversion mesh,with fit statistics and the mesh (InversionAgent)
geo_modelunit and Geo-group labels per cell, their statistics,and the figures (GeoAgent, PetrologyAgent)
draft_report the interpretation and target ranking (ReportAgent)
report_files the reviewed report, {"report_markdown": path};
its existence is what lets the run finish (ReviewAgent)
PyHydroGeophysX.agents.assistants.geosage.workflow module#
GeoSAGE’s run: what “Auto to report” does when GeoSAGE is the assistant.
Follows PyHydroGeophysX.agents.assistants.WORKFLOW_CONTRACT. The loop,
the step events the studio shows, step-by-step approval and questions to the
user all come from PyHydroGeophysX.agents.runtime.entry.drive(); what is
GeoSAGE’s own is the configuration (configure()), the tools
(tools) and the decision prompt (CONTROLLER_PROMPT).
- PyHydroGeophysX.agents.assistants.geosage.workflow.CONTROLLER_PROMPT = 'You are running GeoSAGE, a geological-interpretation workflow from joint gravity and magnetic inversion. Choose the single next action.\n\n{transcript}\n\nActions you can take now:\n{menu}\n\nReply with JSON only - no prose, no code fence. Give your reasoning first:\n{{"why": "<one or two sentences: what the run has shown so far, and so what to do next>", "tool": "<name from the list>"}}\nor, when the reviewed report exists:\n{{"why": "<one short sentence>", "done": true}}\n\nRules:\n- Choose only from the list above. A tool not listed cannot run yet, usually\n because something it needs has not been produced.\n- Compile the geological priors before building the quasi-geological model when\n the user supplied geological information; without priors the grouping falls\n back to GMM/BIC, and the report must say so.\n- The report is final only once it has been reviewed.\n- If a step failed, do not repeat it unchanged; move on and let the report state\n what is missing.\n'#
How the controller is told what kind of run this is.
{transcript}and{menu}are filled in by the controller.
- PyHydroGeophysX.agents.assistants.geosage.workflow.configure(payload: Dict[str, Any]) Dict[str, Any][source]#
The run’s configuration from the request: GeoSAGE’s ContextAgent.
Placeholder for the port: today it only carries the request and the input files through. The ContextAgent turns the request into a complete executable configuration - data fields, inversion settings, grouping settings - on top of the defaults.
- PyHydroGeophysX.agents.assistants.geosage.workflow.run(payload: Dict[str, Any], progress: Callable[[...], None], *, approve: Callable[[...], Any] | None = None, on_event: Callable[[Dict[str, Any]], None] | None = None, events: list | None = None, **_hooks: Any) Dict[str, Any][source]#
Run GeoSAGE on
payloadand return the studio’s result.
Module contents#
GeoSAGE: geological reasoning from joint gravity and magnetic inversion.
Geological Semantic Analysis for Geophysical Exploration (Fang et al., in review; source at ZhengyangFang/GeoSAGE, MIT licence). A user states an exploration objective; GeoSAGE prepares the gravity and magnetic data, compiles geological and petrophysical priors, runs a deterministic joint inversion (SimPEG, cross-gradient coupled), turns the density and susceptibility volumes into a quasi-geological model (prior-guided units and Geo groups, or GMM/BIC when priors are thin), ranks targets, writes the interpretation and has it reviewed before it is final.
Status: being ported. This package is the scaffold the port fills in. The
pipeline is laid out as tools in tools - one per stage of the paper’s
Figure 1, with what each requires and produces - and workflow runs
them on the same controller as AQUAH, so the studio’s chat, live timeline,
step approval and report view work for GeoSAGE as soon as the tools do. Until
then it is listed in the studio’s assistant menu but cannot be selected; set
status="ready" below once the tools run.
The paper’s seven agents map onto this package as follows:
ContextAgent |
|
DataAgent |
|
PetrologyAgent |
|
InversionAgent |
|
GeoAgent |
|
ReportAgent |
|
ReviewAgent |
|