Run history
See what every run read, concluded, proposed and changed, and undo a change.
Every time an agent or rule runs, Meerkats records the whole run: what it read, what it concluded, what it proposed, who decided, what changed, and what happened next. That record is the trace. It is how you check an agent’s work before trusting it, undo a change, and see whether a proposal was right. Nothing an agent does is unrecorded.
Open Automation Agents and select the Agents run history tab.
The run history page#
The counters at the top summarise the workspace’s agents: healthy, needs review, critical, paused and scheduled. Below them, filters:
- Platform: Meta, Google, Blinkit & Zepto, Flipkart, Amazon. Platforms you haven’t connected show not connected.
- Cadence: from event-based and every 5 minutes up to monthly and once.
- All runs, Prod or Test: test runs read real data but send and stage nothing.
Runs are grouped by cadence, and the page explains the grouping: Daily is triage and protect (fast signal, mostly reversible, automatable); Weekly is optimise (enough data for significance; recommend, then approve); Monthly is strategise (structural, slow-signal; the engine frames, a person decides). Paused rows keep their history but don’t run.
Refresh reloads the list. Select a run to open it.
What a run records#
| Section | What it holds |
|---|---|
| Trigger | What started it: the schedule, a rule firing, a watch diverging, a request in the chat, or Run now, and when. |
| Inputs | Every source it read, with the sync time of each. Stale sources are marked. |
| State | What it established: which entities, since when, what changed. With the checks that ran (freshness, lag, dedupe) and their results. |
| Investigation | The path it took through the relationship graph, and what it found at each step. |
| Decision | What it concluded, the evidence, and the why-chain. Or: inconclusive, with what would decide it. |
| Proposal | Each line: action, target, before state, the change, risk, forecast with its prior, reversible, expiry. |
| Approval | Who approved, edited or rejected, when, and any reason given. Or: ran alone, under which autonomy setting. |
| Execution | The live re-read, the write, the read-back, and the value replaced. Or the exact step that failed and the platform's error. |
| Watch | Predicted versus actual at each read, and the verdict: on track, diverged, or measurement break. |
| Credits | What the run used. |
Every number in the run carries its source label: computed, live, yours, your data, and so on. One click on a value opens the query or the platform read behind it. A number without a source never appears. See Where every number comes from.
Undo a change#
- Open the run
Find it in Agents run history, or from the proposal in your Inbox after it was approved.
- Select the executed line
It shows the value it replaced (for example the previous budget) and the rollback that restores it.
- Roll back
Confirm. Meerkats re-reads the target live, restores the stored value, reads it back, and records the reversal as its own run with you as the actor.
Judge an agent on its record#
The run history is how autonomy is earned. For each agent and action type, it accumulates:
- How many proposals were approved, edited or rejected, and the reasons given.
- Predicted versus actual for every executed change: did the metric move as forecast, within the band, over the window?
- How often a run ended inconclusive, needed review, or hit stale data.
Meerkats recommends switching an action type to auto only when this record clears an accuracy bar over enough executed actions. You can see the record on the agent’s page before you decide. See Autonomy.
Inbox Activity#
The third tab on the Agents page, Inbox Activity, is the same history seen from the Inbox’s side: every proposal and suggestion sent, with its decision and outcome. Use it to answer “what did we approve last week, and did it work?”
The workspace Activity log#
Run history is per agent. The workspace Activity log is the account-level record: sharing and access changes, data changes, and automation, each with the person or app that did it and the time. It also logs every API call made against the workspace. Open it from Workspaces in your account menu, then the share icon on the workspace; filter by All activity, Access & sharing, Data changes or Automation. See Governance and audit.
Run history over the API#
GET /agents/run-history
→ recent runs across the workspace's agentsBoth need a key with read scope. Runs can also be exported from the run history page as CSV or JSON for your own records.
Troubleshooting#
No runs yet
The page says: test or enable an agent on the Agents page and it will show up here. A new workspace has no runs until an agent runs.
A run says data stale
The source it needed hadn’t synced recently. Check the platform in Data Hub › Bronze Layer and select Resync; the agent re-runs on its next slot.
A run says inconclusive
The evidence didn’t support a confident answer. The run names exactly what data or time would decide it. That is the correct behaviour, not a failure.
I can't find a run from last month
Run history keeps every run for the life of the workspace. Widen the cadence and platform filters, and check All runs is selected rather than Prod or Test.