Agents#
pycsamt.agents provides small, composable agents for interpreting a
request, planning a workflow, running scientific tools, and coordinating
multi-step processing. The examples in this section run entirely offline:
each one executes inside
AGENT_CONFIG.offline(), so the
agents use deterministic rule-based logic and local numerical code — no API
key, no network request, and zero cost — even when provider keys are present in
the environment.
The sequence progresses from the smallest building blocks to the high-level
AgentMaster front door, and every example ends with a
figure:
route a plain-language request to an intent (confidence chart);
extract a structured workflow configuration (parsed-config table);
inspect the offline model catalogue (architecture/depth chart);
run a local 1-D forward model (earth model + synthetic sounding);
validate a workflow plan before execution (pass/fail report card);
preview a coordinated multi-agent chain (flow diagram);
plan a complete workflow through
AgentMaster(pipeline diagram).
