Agent Observability
Observability in Codebolt is how you see what an agent did, what it's doing right now, and why it made a particular choice. It spans running processes, historical runs, and LLM-level traces, with execution replays linked from Code Observability.
Panels in this section
| Panel | Purpose |
|---|---|
| Agent Debug | Per-instance log stream: LLM calls, tool calls, status events, errors — with filters and history mode |
| AI Debug & Console | Raw LLM inputs/outputs, token counts, latency, cost estimates, plus agent stdout/stderr console |
Related observability panels
Other views that live alongside the agent lifecycle:
| Panel | Location |
|---|---|
| Running Agents (currently executing processes) | Agents → Running Agents |
| Background Agents (trigger-started, detached runs) | Agents → Running Agents |
| Thread Panel (conversation and run history) | Agents → Thread Panel |
| Narrative Graph (visual run replay) | Code Observability → Narrative Graph |
When to reach for which
- Something is still running and you want to watch it → Running Agents
- An agent started itself (cron, webhook, file-change trigger) → Background Agents
- You want to see why the agent did what it did → Narrative Graph
- You want the exact prompt the LLM received → AI Debug
- You want the full per-instance log stream → Agent Debug
Related
- Guardrails & Settings — constrains what agents can do
- Debug & Observability — environment, plugin, and problems panels
- Code Observability — codemap, narrative graph, and structural execution views
- Execution & Tracking — jobs queue and chat canvas