Directory / Pydantic AI

Pydantic AI

Agent framework · Pydantic · 5/7 verified · Last checked 2026-10-16

requires_approval and ApprovalRequired pause tool calls for a human. Toolsets can filter which tools are visible each step. Commands can run in Modal or E2B sandboxes.

Overall: B (2.00 avg, 5/7 controls verified)

Lock this down first

Mark write tools requires_approval=True and run the coding harness in an E2B or Modal sandbox.

Scorecard

Criterion	Rating	Notes
Permission granularity	2/3	FilteredToolset decides per step which tools are available, based on run context.
Human approval gates	2/3	requires_approval=True or raising ApprovalRequired defers the call; you approve or deny it before the run resumes. Nothing admin-lockable.
Sandboxing	2/3	Off by default. Opt-in Modal and E2B sandboxes run commands and file edits off your machine.
Secrets handling	Unverified	Not confirmed in the docs we checked.
Audit logs	2/3	Logfire / OpenTelemetry instrumentation emits spans for each model request and tool call.
Kill switch / interrupt	2/3	A CancellationToken cancels a run in flight and raises RunCancelled with the history so far.
Network egress control	Unverified	Not confirmed in the docs we checked.

Unverified: Secrets handling, Network egress control. That means we could not confirm it in public docs, not that it is missing.

Sources

Vendor? Request a correction. We review corrections within 7 days.