Execution log
Status
Stable
The Execution log is the audit feed of every action the AI agents performed on behalf of your workspace. Every tool call, with its arguments, return value, and timing, is recorded here.
What you see
┌───────────────────────────────────────────────────────────────────┐
│ Filter: Agent ▾ Tool ▾ Date ▾ User ▾ │
├───────────────────────────────────────────────────────────────────┤
│ 2026-05-29 09:14:22 Finance GetGLBalance 387ms ✓ │
│ account=4111, date=2026-05-29 │
│ → {"balance": 142567.40, "currency": "BGN"} │
│ │
│ 2026-05-29 09:14:23 Finance GetCashFlow 210ms ✓ │
│ period="this_week" │
│ → [29 data points, total +8200] │
│ │
│ 2026-05-29 09:11:08 Warehouse GetReorderAlerts 145ms ✓ │
│ → [{"sku":"SKU-2104", ...}, {"sku":"SKU-2106", ...}, ...] │
│ ... │
└───────────────────────────────────────────────────────────────────┘Each row:
| Column | Detail |
|---|---|
| Timestamp | When the call started |
| Agent | Which agent initiated the call |
| Tool | The function called |
| Latency | Time the tool took to return |
| Status | ✓ success / ✗ error |
| Arguments | The call's input (redacted for sensitive fields) |
| Return | The call's output (truncated for large results) |
Click any row to see the full untruncated detail.
What's recorded
- Read tool calls — every fetch from your tenant Odoo
- Would-execute (Shadow Mode) — simulated write actions, logged but not applied
- Workspace API key calls — calls authenticated by a
pa_live_…key - Errors — failed tool calls with the error message
What's not recorded:
- Chat messages themselves (those live in the conversation, retained separately for 7–90 days per the workspace's chat retention)
- Browser page views / clicks unrelated to a tool call
- Internal platform telemetry
Why this exists
Three reasons:
- Trust — if you ever wonder what an agent actually did, the answer is here, byte-for-byte.
- Debugging — when an agent gives an unexpected answer, the tool calls show you exactly what data it saw.
- Compliance — auditors can see the full trail of automated actions over the past 90 days.
Retention
| Where | How long |
|---|---|
PLANA:executions:{workspace} Redis | 30 days |
| Loki audit log (PLANA-side) | 90 days |
| Compliance archive (Enterprise tier) | 7 years on request |
The 30-day Redis log is what the BOS view shows. Older entries are available on request from PLANA via the workspace Matrix room.
Filtering
| Filter | Use |
|---|---|
| Agent | Just Finance, just Sales, etc. |
| Tool | Specific tool name (GetGLBalance) |
| Date range | Pick start / end |
| User | Show only what one team member triggered |
| Status | Only failures |
| API key | Calls made by a specific pa_live_… key |
Combine filters. The URL preserves the filter so you can bookmark "my failed Finance calls last week".
Exporting
Click ⇩ Export to download the visible entries as JSON or CSV. Useful for offline analysis or attaching to a support ticket.
The export respects the active filters — so "export failures" gives you just the failures.
Sensitive-field redaction
Some fields are redacted before being shown here:
| Field type | Redaction |
|---|---|
| IBAN | Last 4 digits only |
| Tax ID full | Last 4 only |
| Passwords / tokens | "REDACTED" |
Redaction is consistent with what the agent sees — the LLM never sees the unredacted value, and neither does the audit log.
Where to read more
- How agents work — what generates these entries
- Safety and limits
- Settings → API keys