You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 4705345
Browse filesBrowse the repository at this point in the historyBrowse files
Both applications pin the shared agent runtime to an exact Git revision in their Cargo manifests and lockfiles. Each application owns its permission policy and database-operation classification.
16
16
17
17
The application-owned `ai_permissions` module resolves policy, migrates legacy settings, classifies database payloads, and gates both MCP and built-in agent execution. The shared agent routes requests through an application preparation hook and carries its confirmation risk and context; it does not resolve this permission chain. Execution rechecks the application policy on every attempt.
18
+
19
+
AI schema reads use the same application gate as tool calls. Cached schema and editor context are included only when the associated connections permit reading. Manual UI capability invocations retain their existing access behavior.
20
+
21
+
Agent executors are bound to a conversation and its connection/source identities. Before dispatch they reload connection policy, current attachment restrictions, and confirmation rules. Detaching or deleting a source prevents subsequent dispatches. The sidebar uses its bound active connection without requiring a Data Studio attachment.
22
+
23
+
Preparation rejects denied calls before displaying confirmation. A queued call whose confirmation requirements become stricter returns to the existing approval flow. Permission denials are terminal and do not trigger transport retries; approval never bypasses a later denial. Rechecks occur at dispatch and do not cancel operations already dispatched or erase context already sent to a model.
0 commit comments