Auditability
Your agent did 43 things last night. Can you name one?
Agents work fast. That's the point. But speed without visibility is just a faster way to break things. Civic logs every tool call, every parameter, every response. When something goes wrong at 3am, you get a full replay instead of a shrug.
scheduling-agent / 2026-03-03 19:28Z
recording
#Time (PST)ToolServiceStatusDuration
0 / 7 actions
Three steps. Zero instrumentation.
Civic Chat
1[you] What did review-bot do yesterday?
2
3[nexus] review-bot · 2026-03-03 · 6 calls
4 02:14 gmail:list_emails — ok 120ms
5 02:14 gmail:read — ok 85ms
6 02:14 github:create_pr — ok 340ms
7 02:14 slack:post — ok 95ms
8 02:14 postgres:query — ✗ blocked
9 02:14 nexus:policy — warn 3ms
What you actually get
Full action history
Every tool call is recorded with timestamp, tool, service, status, and duration. Nothing gets lost.
Single-tenant, ephemeral MCPs
No chats stored. This prevents data or context from leaking between users or requests.
Tool call forensics
Inspect every tool invocation: inputs, parameters, and outputs. Understand exactly what an agent did and why.
Every action. Accounted for.
Start logging what your agents actually do.