Version 1.2.2. Added usage overview. Shows token used and cost in $.

server/agents/tasks.py

@@ -1,5 +1,20 @@
 """
-agents/tasks.py — Agent and agent run CRUD operations (async).
+agents/tasks.py — Agent and agent run CRUD operations (async, PostgreSQL).
+
+Agents are persistent, named, goal-oriented configurations that can be:
+  - Run manually (via the Agents UI or /api/agents/{id}/run)
+  - Scheduled with a cron expression (managed by AgentRunner + APScheduler)
+  - Triggered by email, Telegram, webhooks, or monitors
+
+agent_runs records every execution — input/output tokens, status, result text.
+This is the source of truth for the Agents UI run history and token totals.
+
+Design note on allowed_tools:
+  - NULL in DB means "all tools" (unlimited access)
+  - [] (empty list) is falsy in Python — treated identically to NULL
+  - Non-empty list restricts the agent to exactly those tools
+  - This is enforced structurally: only declared schemas are sent to the model;
+    it is impossible for the model to call a tool it wasn't given a schema for.
 """
 from __future__ import annotations
 
@@ -161,6 +176,8 @@ async def finish_run(
     output_tokens: int = 0,
     result: str | None = None,
     error: str | None = None,
+    model: str | None = None,
+    cost_usd: float | None = None,
 ) -> dict | None:
     now = _now()
     pool = await get_pool()
@@ -168,10 +185,12 @@ async def finish_run(
         """
         UPDATE agent_runs
         SET ended_at = $1, status = $2, input_tokens = $3,
-            output_tokens = $4, result = $5, error = $6
-        WHERE id = $7
+            output_tokens = $4, result = $5, error = $6,
+            model = $7, cost_usd = $8
+        WHERE id = $9
         """,
-        now, status, input_tokens, output_tokens, result, error, run_id,
+        now, status, input_tokens, output_tokens, result, error,
+        model or None, cost_usd, run_id,
     )
     return await get_run(run_id)