Limits and safety
Every limit SOAR enforces, which ones you can change, how runs behave during an alert storm, and how to keep automation safe.
Plan your flows around the limits below. nano enforces each one; a flow or request that exceeds a limit is refused with a message that names it. Most limits are fixed. The ones you can change say how.
Runs and execution
| Limit | Value | Can you change it? |
|---|---|---|
| Runs executing at once | 8 for each jobs service process | Fixed |
| Vendor requests in flight at once | 16 for each process | Yes: SOAR_MAX_CONCURRENT_EXCHANGES, 1 to 256 |
| Wait for a free request slot | Up to 10 seconds for a single action. A flow step doesn't wait. In both cases the run stays queued and tries again. | Fixed |
| Time for one action step | 15 minutes | Fixed |
| Time for one vendor request | 1 to 60 seconds, set by the pack | Set by the pack |
| Vendor requests per pack | Set by the pack, up to 600 a minute and 200,000 a day | Set by the pack |
| Attempts of one step | 32 | Fixed |
| Retries of one action | 1 to 5. Constant delay of 1 to 300 seconds, or a growing delay from a 1 to 60 second base with a multiplier of 1 to 300. | Per step, within these bounds |
| Run inputs | 256 KiB | Fixed |
| Recorded output of one action | 256 KiB. Larger output is recorded as truncated. | Fixed |
| Recorded output of a run | 240 KiB. Larger output is recorded as truncated. | Fixed |
SOAR_MAX_CONCURRENT_EXCHANGES is read from the environment of the nano API and jobs services when they start. A value outside 1 to 256, or one that isn't a whole number, falls back to 16.
Flows
| Limit | Value | Can you change it? |
|---|---|---|
| Flows | 500, not counting archived flows | Fixed |
| Steps in a flow | 64 | Fixed |
| Nesting depth | 4 levels | Fixed |
| Items in one loop | 100 | Fixed |
| Loop items in one run | 100, across all loops | Fixed |
| Loop order | Items always run one at a time, in order. From v2.0.33 the editor no longer offers parallel loops. | Fixed |
| Parallel branches | Not supported. A flow with them fails validation. | Fixed |
| Flow definition | 1 MiB and 20,000 values | Fixed |
| Triggered runs per flow (from v2.0.33) | 60 per rolling hour by default. Extra triggers are recorded as Skipped · rate limited and kept 7 days. Manual runs don't count. | Per flow, up to 1,000, under Limits |
| Circuit breaker (from v2.0.33) | Pauses the flow after 5 failed triggered runs in a row, or above 600 triggers in an hour. A run whose approval expired unanswered doesn't count as a failure. | Spike threshold per flow, under Limits. Resuming needs soar:manage and a reason. |
| Triggers held while paused (from v2.0.33) | Up to 100, replayed oldest first after you resume the flow | Fixed |
Triggers
| Limit | Value | Can you change it? |
|---|---|---|
| Runs per event | One run of each matching flow for each alert, case event, or triage verdict | Fixed |
| Schedule interval | At least 5 minutes between runs | Per flow, within this bound |
| Event match | Up to 8 clauses and 8,192 bytes | Fixed |
| Rules in an alert trigger | Up to 64 rule IDs | Fixed |
| Trigger queue | 20,000 events waiting. New events beyond that are dropped and counted. | Fixed |
| Retries of a trigger event | 10 attempts, waiting 5 seconds and doubling up to 5 minutes, then the event is set aside (dead-lettered) | Fixed |
| Dead-lettered events kept | 7 days, newest 10,000 | Fixed |
Steps
| Limit | Value | Can you change it? |
|---|---|---|
| Search time window | Up to 30 days | Per step, within this bound |
| Search rows | 1 to 1,000 | Per step, within this bound |
| Search time | 30 seconds | Fixed |
| Search query | Up to 64 KiB | Fixed |
| Search output | 240 KiB. Rows beyond that are left out and the output is marked truncated. | Fixed |
| Judge token cap | 2,000 by default, up to 8,000 per call | Per step, within this bound |
| Judge evidence | 1 to 32 values, 32 KiB together | Fixed |
| Judge labels | 1 to 32 | Fixed |
| Judge timeout | 60 seconds | Fixed |
| Judge cache | 6 hours by default, up to 24 hours | Per step, within this bound |
| AI Judge daily token budget | 250,000 by default, 0 to 1,000,000,000 | Yes: Settings → SOAR |
| Approval wait | 1 minute to 24 hours for an approval gate (default 24 hours). 24 hours for an action or notification. | Per approval gate, within this bound |
| Approval choices | Up to 32 | Fixed |
| Approval reason | Up to 2,048 characters | Fixed |
Notifications
| Limit | Value | Can you change it? |
|---|---|---|
| Notifications sent | 200 in any rolling 24 hours, email and webhook together, including test sends | Fixed |
| Email recipients | 50 | Fixed |
| Webhook endpoints | 50 | Fixed |
| Subject | 200 characters | Fixed |
| Body template | 8 KiB. Rendered body up to 16 KiB. | Fixed |
| Values bound into a notification | 16, each up to 200 characters | Fixed |
Connections and outbound traffic
| Limit | Value | Can you change it? |
|---|---|---|
| Connections | 256 | Fixed |
| Credential fields | 64 per connection, 16 KiB each | Fixed |
| Private-network allowlist | 64 entries | Fixed. You edit the entries on Settings → SOAR. |
| Outbound destinations | Public addresses only, pinned for the run. Private hosts only through the allowlist, for packs marked for self-managed tools. | Allowlist only |
| Redirects | Not followed | Fixed |
| Air-gapped catalog age | 30 days by default | Yes: SOAR_OFFLINE_INDEX_MAX_AGE_DAYS, 1 to 90 |
| Vendor calls per connection (from v2.0.33) | 60 per minute by default, across every flow. A step that finds the minute full waits for the next minute. | Per connection, up to 600 |
How runs behave under load
During an alert storm, SOAR works through runs in order and doesn't merge alerts. From v2.0.33, a flow's run cap and circuit breaker bound how many runs one storm can start.
- One run per alert per flow. If a rule fires 1,000 alerts and two live flows match each one, nano starts 2,000 runs.
- Runs queue. Each jobs service process executes up to 8 runs at once. The others wait in the queue and start oldest first as workers free up. A run that waits for approval doesn't hold a worker.
- Vendor requests share a pool. At most 16 requests to vendors are in flight per process. A flow step that finds no free slot goes back to the queue and tries again.
- Packs have budgets. When a pack's request budget for the minute or day is used up, its steps fail with
rate_limitedinstead of piling up requests. - The trigger queue has a ceiling. If 20,000 events are already waiting, nano drops new trigger events and counts them. The Flows page shows Trigger events dropped with the count and the last time. Alerts and cases themselves are always saved.
- Approvals pile up too. Each run that needs approval creates its own approval.
From v2.0.33, each flow has a triggered-run cap and a circuit breaker (see Flows). Flows and connections that existed before the v2.0.33 upgrade show No limit (set before upgrade) for the caps until you tick Apply the default limits; the failure breaker applies to every flow. Scope your triggers anyway, so a storm can't flood your queue or your approvers.
Keep automation safe
- Scope triggers tightly. Filter alert triggers by severity and rule, and add an event match on the entities you care about. Severity and rule filters are checked as the alert is written, so an alert that no live flow wants never enters the trigger queue. See Triggers.
- Gate disruptive actions. Put an approval gate, or Always require approval, in front of anything that contains hosts, suspends users, or blocks traffic. Destructive actions always wait for a person.
- Require approval on notifications you want to review. From v2.0.33, an email step waits for approval unless you choose Send without approval. Webhook steps send without approval unless you set
approval: required. - Test before you activate. A test with a real alert shows what each step would do without changing anything. See Test a flow.
- Use a dedicated owner. A flow acts for the person who last saved it. Save production flows from an account whose permissions match what the flow should do, and have someone else approve.
- Prefer actions that are safe to retry. nano resends an action after an unclear outcome only when the pack marks it safe to retry. Otherwise a person decides.
What's next
AI Judge
Ask an AI model a fixed question about evidence a flow already holds, branch on its labeled verdict, and control what it costs with token caps, caching, and a daily budget.
Audit and troubleshooting
Find what SOAR did and why: audit events, run statuses, failure reasons, and fixes for flows that don't fire, runs that fail, and dropped trigger events.