nano SIEM
Releases

nano v2.0.33

Supported in:Self-hosted

nano v2.0.33: an nginx security update, run caps and a circuit breaker for SOAR, team and private escalation queues and a duplicate gate for Hunts, one risk score per entity, and search guards based on query cost.

v2.0.33 includes everything in v2.0.32. It includes security fixes: the nginx edge moves to 1.31.9, which fixes five nginx CVEs, HTTP/2 is turned off on the Compose TLS edge, and SOAR email gets fixes for six SMTP advisories. We recommend that every self-hosted install upgrades.

To upgrade, follow Upgrade a self-hosted deployment and the Upgrade notes below. v2.0.20 is a required stop: installs on a release older than v2.0.20 upgrade to v2.0.20 first.

Security

  • nginx 1.31.9. The nginx image that serves the nano edge moves from 1.31.8 to 1.31.9 in every edition: Compose (Open and Self-run), role-split Compose and Helm. 1.31.9 fixes CVE-2026-123007 (HTTP/2 flow-control memory corruption), CVE-2026-123633 (HTTP/2 use-after-free), CVE-2026-121810 (HTTP/3 use-after-free), CVE-2026-123634 (resolver use-after-free) and CVE-2026-118899 (TLS 1.3 request body infinite loop).
  • HTTP/2 is off on the Compose TLS edge. The Compose TLS edge was the only nano nginx configuration with HTTP/2 turned on. It now serves HTTP/1.1. Nothing in nano needs HTTP/2 through nginx: OTLP over gRPC goes to Vector's port 4317 directly. Browsers and clients fall back to HTTP/1.1 on their own. The upgrade applies the new configuration for you.
  • SMTP email fixes (Self-run). The email library behind SOAR approval and notification email moves to lettre 0.11.24, which fixes six SMTP advisories (RUSTSEC-2026-0560 to RUSTSEC-2026-0565), including STARTTLS reply injection and missing dot-stuffing.

New

Hunts and SOAR (Automation) now carry a BETA badge in the left rail and on their page titles. Both are included with nano Self-run and may change based on feedback.

SOAR: run caps, budgets and a circuit breaker

A noisy rule can no longer queue thousands of runs of every alert-triggered flow. See Limits and safety.

  • Per-flow run cap. A flow runs at most 60 triggered runs per hour by default (up to 1000). Extra triggers are recorded as Skipped · rate limited and are kept for 7 days. Manual runs don't count.
  • Per-connection call budget. A connection makes at most 60 vendor calls per minute by default (up to 600), across every flow. A step that finds the minute full waits for the next minute, and nothing is sent until there is room.
  • Circuit breaker. A flow pauses after 5 failed triggered runs in a row, or above 600 triggers in an hour. A paused flow shows Paused by breaker on the Flows list and a banner on its page, and SIEM Health raises a finding. Resuming needs the soar:manage permission and a reason, and is audited. A run whose approval expired unanswered doesn't count as a failure.
  • A paused flow keeps its triggers. While a flow is paused, up to 100 matching triggers are held and run, oldest first, after you resume it. The flow page shows how many are held.
  • Existing flows and connections are exempt from the default caps. Flows and connections that existed before the upgrade aren't limited by the new default caps and budgets until an admin opts in. The failure breaker applies to every flow. See SOAR limits after the upgrade.
  • Email steps wait for approval by default. A new email step asks an approver before it sends unless you choose Send without approval. Email steps in flows saved before the upgrade keep sending as before. See SOAR approvals after the upgrade.

SOAR and Hunts work together

  • A hunt escalation trigger. A flow can start when a hunt escalation reaches the main queue, and match on the hunt, severity, entity type and the reviewer's confidence.
  • Set hunt finding disposition. A new native step records a ruling on the hunt finding (true positive, benign, false positive, duplicate or needs more). Later hunt runs see it in their memory, marked as set by a flow. It doesn't count as an analyst's ruling or toward the reviewer's measured precision.
  • Hunt-triggered runs need a person. On a flow triggered by a hunt escalation, every pack action, notification and disposition step waits for a human approval, whoever saved the flow.

Hunts

  • Team and private escalation queues. Each hunt now chooses where an escalation lands: the main queue, a team queue, the hunt owner's private queue, or nowhere. A team or private escalation is held until a team member, the owner or an admin promotes it to the main queue or dismisses it. A team queue notifies only the team's own channel. The Escalations page has Main, My team and Private tabs. See Escalation.
  • A duplicate gate. A finding with the same shape as an open finding from the same hunt within the window (an unreviewed lead, or an escalation nobody has dismissed or closed), (7 days by default, 0 to 90 per hunt) is stored as a duplicate and costs no reviewer tokens. A worse repeat, such as a higher severity or a success after failures, is never gated. A repeat of an open finding raises that finding's score. The gate is off for hunts that existed before the upgrade. See Hunts after the upgrade.
  • A per-hunt escalation floor. A hunt can set the lowest verdict it escalates (low, medium, high or critical). Below the floor the finding stays a lead. Identical repeats of a lead the floor refused are gated, so the reviewer doesn't review the same thing every run.
  • A hunt charter. Each hunt can have a charter, with an optional install default: scope, crown jewels, known-good infrastructure, what clears the escalation bar, and analyst notes. The investigator and reviewer read it as guidance, never as permission. Every save is a new version.
  • Stock hunts stay current. Hunts imported from the content channel that you haven't customized are refreshed in place on the next sync. Your schedule, budgets, escalation settings and suppressions are kept. A customized hunt shows Update available, with a dialog to review and apply the update. Re-importing an archived hunt restores it instead of creating a copy.

Improvements

  • Search guards based on query cost. Unfiltered searches with eventstats, streamstats, transaction, anomaly or dedup now run when the estimated number of events fits this install's budget, instead of being refused on sight. stats by a high-cardinality field, values(), list() and reverse always run, because they spill to disk or are capped. A refused search now says how many events it would read and how many this install admits, so you know how far to narrow it.
  • Temporary disk is capped. A large GROUP BY or ORDER BY that spills to disk can no longer fill the ClickHouse disk and stop ingestion. The caps are sized from ClickHouse's memory and, where nano knows it, its disk. Helm sizes them from clickhouse.storage.size.
  • SQL to nPL handles computed columns. Queries an agent runs as SQL with computed columns, expressions over aggregates, computed filters or GROUP BY 1 now convert to nPL instead of showing raw SQL.
  • Risk entity types. Risk entities take their type from the rule's risk entity field, so a hostname without a dot is now a hostname, not a user. One value seen by several rules is one entity with one score, and its type is chosen by a fixed order (email, cloud account, hash, IP, hostname, user). See Search results can change.
  • Hunt definition changes are audited. Creating, archiving, triggering and editing a hunt is audited, and an edit records old and new values for its schedule, budgets and escalation, and which pivots changed.
  • Vector only reloads when its configuration changes. An API restart no longer rewrites Vector's configuration and reloads Vector when nothing changed. Changed files are written atomically.
  • Operators render as typed. !=, <=, >= and == show as typed everywhere, instead of as a single symbol such as ≠.

Fixes

  • Prevalence on wide searches. * | prevalence <field> over several days on OCSF no longer runs out of memory. The search now reads only the page you see. A wide prevalence search that would still run out of memory is refused up front with an event count.
  • Rarity on OCSF proxy and network data. Search → Rarity marked almost every domain RARE, because it counted the proxy appliance as the only host. It now counts client IPs the same way the prevalence summaries do. A domain missing from the baseline is no longer RARE on its own.
  • Later stages on OCSF. On OCSF, | prevalence followed by timechart or stats, sort then rename, numeric bin, and append with rename no longer fail with "Field … does not exist".
  • Joins and lookups use less memory. A join whose main side returns events pages them like a normal search. | inputlookup of a missing table is refused before anything is scanned, and an inputlookup without key= no longer reads events it would discard.
  • Detection health is critical only for rules failing now. A rule that failed and then recovered no longer keeps detection health critical, and lowers the SIEM Health score, for 24 hours.
  • Agent detection switches are enforced everywhere. Turning the install switch off pauses every live agent detection at once, and nothing can turn one back on while a switch is off. The triage budget for agent detections is a hard limit.
  • Hunt runs where the model never called a tool fail instead of ending as a clean run with no findings.
  • nano-jobs on 8 GB hosts. Loading the MITRE ATT&CK catalog no longer crash-loops nano-jobs on 8 GB hosts. The catalog loads in a fraction of the memory, a failed load waits before retrying, and small hosts give nano-jobs at least 384 MB.
  • SOAR runs follow the editor. A flow run still unfinished 24 hours after it started ends as Timed out, and each step runs under its own timeout (at most 15 minutes). Parallel loops are no longer offered, because loop items always ran one at a time.

Upgrade notes

  • Required stop at v2.0.20. See above.
  • Migrations run automatically. New PostgreSQL migrations apply during the upgrade.
  • ClickHouse temporary disk caps. On Compose, nano upgrade adds the caps to your .env, sized for this host's memory and disk, and renders them. On Helm, they follow clickhouse.storage.size, you can override them under clickhouse.profiles and clickhouse.server, and the operator restarts the ClickHouse pods to apply the server-wide cap. On role-split hosts, restart ClickHouse after a rerun that only changes sizing.
  • Vector reloads once. The first API start after the upgrade reloads Vector once. Later restarts reload it only when its configuration changed.

SOAR limits after the upgrade

  • Flows and connections that existed before the upgrade show No limit (set before upgrade) and keep working as before. A value you set yourself still applies.
  • To apply the defaults, open the flow's Limits (or the connection's call budget), tick Apply the default limits, and save. This can't be undone, but you can still raise each limit.
  • New flows and connections get the defaults: 60 triggered runs per hour, a pause above 600 triggers per hour, and 60 vendor calls per minute.
  • The failure breaker (pause after 5 failed triggered runs in a row) applies to every flow, old and new. Failures from before the upgrade don't count.

SOAR approvals after the upgrade

  • The upgrade sets email steps in existing flows to Send without approval, so they keep sending as before. Open a flow to review the choice.
  • A Set hunt finding disposition step on a flow triggered by a hunt escalation waits for approval.
  • Runs already older than 24 hours when you upgrade (waiting in the queue or for an approval) end as Timed out on the first pass after the upgrade.
  • A flow saved with a parallel loop keeps running its items in order. Before you can save it again, click Run items in order.

Hunts after the upgrade

  • Users limited to some sources can no longer change a hunt's escalation mode, reviewer switch, default severity, escalation floor, duplicate window or charter. An unrestricted user or an admin makes those changes.
  • The duplicate gate is off for existing hunts. Hunts that existed before the upgrade, and had no window of their own, get a duplicate window of 0 days. To turn the gate on, set Duplicate window under Escalation floor and duplicates in the hunt editor. New hunts use the install default of 7 days, which you can change on the Runners page.
  • Leads the escalation floor refused before the upgrade don't gate repeats until they are reviewed again.
  • Imported stock hunts that you never edited are refreshed on the next content sync. If you edited one, nano shows Update available and never overwrites your edit.

Search results can change

  • A search that the old unfiltered-search guard refused can now run, and one the new estimate refuses says how many events it would read.
  • | cloud must now be the last command in a query. A query with commands after it returns an error, because those commands replaced the fields the cloud view reads.
  • Search → Rarity shows far fewer RARE domains on OCSF proxy data. That is the fix, not missing data.
  • Risk entity types can change. Hosts and hashes that showed as user now show as hostname or hash. Check user threshold overrides, and saved searches and rules on dataset=risk that filter on entity_type="user".

On this page