Slack notifications

Neens can post its insights (new failure modes, regressions, alert rule breaches, pre-production gate failures, judge drift and more) into the Slack channels your teams already watch. Your company, each org and each agent can have its own channels and its own rules, so the returns team hears about the returns agent and the platform team hears about everything critical, without anyone sharing a firehose.

At a glance

How the pieces fit

• A Slack workspace connection is made once, by an admin, with Add to Slack. Every part of your company can then use it, unless the admin who connected it limits it to their own org or agent. You can connect as many workspaces as your teams use.
• A channel (or an incoming webhook) is a place Neens may post. It is owned by your company, an org or an agent. The owner decides whether rules further down may use it too.
• A routing rule belongs to your company, an org or an agent, and says which insights to send where: insight types, a minimum severity, optional tags, which agents, and one or more channels.

Rules are additive. Every rule that matches an insight sends it, and Neens de-duplicates by channel: if a company rule and an agent rule both send the same insight to #alerts, it is posted there once.

Who can do what

A rule can only send to a channel owned by the same scope, or to a channel an ancestor scope has shared down (the Let rules for orgs and agents below this scope use this channel option). An agent’s rule can never post to another agent’s channel or to a sibling org’s channel.

Connect a Slack workspace

Neens asks for the smallest set of Slack permissions it needs:

If a later Neens feature needs another permission, the workspace shows Missing permissions and an Update permissions button. Nothing stops working in the meantime; only the new feature waits for your approval. A workspace connected before /neens, previews, DMs and @Neens arrived shows Update permissions until an admin approves them.

Enterprise Grid: install for the whole organization

On Slack Enterprise Grid, a Slack org admin can approve Neens for the whole organization in the Approve the app in Slack step. Neens then shows one row for the organization, labelled <organization> (organization), and every workspace in it can use Neens with no separate install.

• Requests from any workspace of the organization reach your Neens account.
• An organization and the workspaces in it belong to one Neens company. If another Neens company already connected the organization, or any workspace in it, the install ends with This Slack workspace is already connected to another Neens company, the same as for a single workspace. The same applies the other way round: a workspace inside an organization that another company connected can’t be connected on its own.
• Slack lists channels one workspace at a time, so when you add a channel from an organization, Neens first asks Workspace in this organization. It lists only the workspaces the app has been added to. If it lists none, or the one you need is missing, a Slack organization admin adds the Neens app to that workspace in Slack’s organization settings first. You can also type /neens subscribe #channel <agent> in the workspace.
• Removing the app from the organization in Slack disconnects it in Neens.

Limit a workspace to one org or agent

By default a connected workspace is usable by the whole company. To stop other teams from adding channels from it, open the row’s ⋯ menu and choose Limit to … (the org or agent it was connected from). Let the whole company use it undoes that.

Add a channel

Recipes

Critical insights for one agent, and a weekly digest for the team

“Send every critical insight for the returns agent to #returns-agent-alerts right away, and everything else about it to #team-weekly once a week.”

A critical insight now reaches #returns-agent-alerts immediately. It also matches the digest rule, so it appears in Monday’s digest in #team-weekly as well. The two rules send to different channels, so each channel gets it once.

A Monday summary of every agent for the team

“Every Monday at 09:00 UTC, post one summary per agent to #agent-review: its KPIs, its top failure modes and the fixes that held last week.”

Each Monday, #agent-review gets one message per agent. Changing the day or time later in the week doesn’t send that week’s digest a second time.

One org-wide channel for every agent in an org

“Post every high-severity insight from any agent in the Customer Support org to #support-quality, except the sandbox agent.”

1. Add #support-quality with Owned by set to the org.
2. Create a rule that Applies to the org, keep All agents in scope, and tick the sandbox agent under Except these agents.
3. Set Minimum severity to High, Delivery to Immediately, and Send to #support-quality.

New agents created in the org later are included automatically, because the rule covers the whole org.

To let each agent’s own admin route insights into that channel too, edit the channel and turn on Let rules for orgs and agents below this scope use this channel.

A company-wide “anything critical” channel

Add #neens-alerts owned by your company, then a company rule: Minimum severity Critical, Delivery Immediately, Send to #neens-alerts. Turn on the share-down option on the channel if teams should be able to add their own rules that post there.

An incoming webhook instead of the app

Use an incoming webhook when you can’t install the Neens app in a workspace, or want to post into a channel without inviting a bot.

Neens encrypts the URL and never shows it again; the list shows only its last few characters. To point the destination at a new webhook, edit it and choose Replace URL.

Writing a rule

New rules default to High and Immediately.

Cooldown: held, never dropped

A cooldown limits how often one rule posts to one channel. After the rule sends an insight to a channel, any further insights it matches for that channel during the cooldown are held, then sent together as one summary message when the cooldown ends. Nothing is dropped. Updates to cards that were already posted (resolved, recurred) are never delayed by a cooldown.

Digests

A digest rule holds the insights it matches and posts one summary at the chosen hour (daily) or day and hour (weekly), listing each insight and its current state, so an insight that resolved in the meantime is shown as resolved rather than silently left out.

Weekly agent digest

A Weekly agent digest rule posts one summary per agent, once a week, on the day and hour (UTC) you choose: the agent’s KPIs against their targets, its top open failure modes, and the fixes that held in production that week. It ignores insight types, tags, severity and cooldown, and always sends metadata only. A section Neens could not read says so instead of showing zero. If Neens was down at that hour, the digest still goes out within the next few hours, and never twice for the same week.

Low-severity insights are usually best left to digests, so picking Low for a Matching insights rule switches its delivery to Digest (you can change it back).

When several rules match

Every matching rule counts, and each channel gets the insight once. If any of those rules wants it immediately (and is not in a cooldown), it is posted immediately; otherwise it waits for the earliest cooldown end or digest. Excerpts are included only if every rule sending to that channel asks for them.

What a message looks like

Each insight is posted as a card with its title, a short summary, its severity, the agent and company it is about, a few key numbers, and a link back to Neens. When the same insight recurs, Neens updates the original card and replies in its thread instead of posting a new one; when it is resolved, the card is updated and a reply says so.

For example, an alert rule breach on the Help Bot agent arrives in #agent-alerts as:

┃ Alert: Error rate
┃ Error rate is 25% over the last 1h.
┃
┃ Value        Threshold
┃ 0.25         > 0.1
┃ Measure      Window
┃ error_rate   1h
┃
┃ [ Open alerts ]
┃ High · Open · Acme Corp · Help Bot · 2026-10-14 09:30 UTC

The coloured bar on the left follows the severity (red for critical and high, orange for medium), turns green once the insight is resolved and grey when it is muted. The same card is what you see in a notification preview: [HIGH] Alert: Error rate — Acme Corp · Help Bot · 2026-10-14 09:30 UTC.

A digest or a cooldown summary is one message listing every held insight with its severity, its current state and a link, for example [HIGH] 2 insights since 2026-10-13 09:00 UTC.

With Redacted excerpts turned on for both the rule and the agent, a failure-mode or novelty card also carries a short excerpt of one example conversation (User: … / Agent: …), with personal data masked. See What Neens shares with Slack.

Muting an insight in Neens greys out its card in Slack and posts nothing new. A webhook message can’t be edited, so for a webhook destination a resolve is posted as a short Resolved: … message and a mute posts nothing.

Link your account

Buttons on Neens cards in Slack act as you, so Neens first needs to know which Neens login belongs to your Slack account. You link it once per Slack workspace (and per Neens account, if several Neens accounts share that workspace).

Linked Slack accounts in Settings → Integrations lists the Slack accounts linked to you, with Unlink. A company-wide admin can switch to Everyone to see and unlink any account in the company. Removing someone from the company unlinks their Slack accounts at once.

Act from Slack

A card about a new failure mode (or a novel failing session) shows the failure mode, how many sessions it affects, a link to an example trace, and three buttons:

Each button acts once per press: pressing the same button again within a minute (a double-click) is ignored. Each button runs the same permission check as Neens itself, as you. If your role doesn’t allow an action, Slack tells you privately (Your Neens role doesn’t allow this) and nothing changes. A card from another Neens account in a shared workspace only works for people linked in that account.

Follow a fix in the card’s thread

Once a failure-mode card is posted, the fix’s progress is added as replies in that card’s thread, so one thread tells the whole story:

Replies are added only under a card that was actually posted to that channel. Incoming webhooks can’t be threaded, so a webhook channel gets these as separate messages, without buttons.

The /neens command

Type /neens in any channel or DM of a connected workspace. Replies are visible only to you, and every command runs as you, with your Neens permissions. If your Slack account isn’t linked yet, /neens only replies with how to link it (or type /neens link).

The agent is matched by its name (or its slug); quotes are optional. With no agent, status and failures cover every agent you can open, and subscribe and review ask which one you mean. Neens only ever lists agents you can open, so a name you can’t access gets No agent called … that you can open in Neens. Use here instead of #channel to subscribe the channel you are typing in. If Neens isn’t in that channel yet, the reply tells you to type /invite @Neens there.

Link previews

When someone posts a link to Neens in Slack, Neens adds a preview under it:

A preview is visible to everyone in the channel, so it never includes trace content, even for agents that allow excerpts in cards. Links need the agent in them to get a full preview, as links copied from Neens do (…/clusters/<id>?project=<agent>). Links to other sites are left alone.

Your morning inbox DM

Neens can DM you each morning with the items that need you, ranked like your action inbox in Insights: each item’s title, severity, agent and how many sessions it hit in the last 24 hours, plus a link. It lists only agents you can open, and it is sent only on days when something needs you.

It is off until you turn it on, for each linked Slack account:

• in Neens: Settings → Account → Slack, switch Morning inbox DM on and pick Send at (shown in UTC and in your own time), or
• in Slack: /neens inbox on 8 (8:00 UTC), and /neens inbox off to stop.

The DM arrives in the Neens app’s Messages tab in Slack.

Review traces from Slack

Label traces without leaving Slack: type /neens review <agent> and Neens DMs you the next trace from your review queue for that agent, with why it was picked and Pass and Fail buttons. Your verdict is saved as your label, exactly as on the Review page, and is what judges are aligned against. The next trace follows automatically until your queue is done.

• The DM shows a short, redacted excerpt of the trace (the user’s first message and the agent’s final answer, with personal data masked) only when the agent’s Data shared with Slack setting is Redacted excerpts. Otherwise it shows the reason and a link to the trace.
• Reviewing needs the member or admin role on the agent. A viewer gets Your Neens role can’t review traces in ….

Ask @Neens in a thread

Mention @Neens with a question, in a channel Neens is in, and it answers in the thread:

@Neens what’s failing in Returns Bot this week?

@Neens which fixes held after release for the Stylist agent?

@Neens how do I set up a weekly digest?

• It answers as you, from the Neens data you can read, using the same tools as the assistant in Neens, but read-only: from Slack it can’t create, change or run anything.
• It uses the LLM connection your agent uses in Neens. If none is set up it says so; an admin can add one in Settings → LLM providers.
• Ask about an agent you can’t open and it refuses, without reading anything about it. Refusals and setup hints are shown only to you, not to the channel.
• Links in its answers are live only when they point to Neens; anything Slack would treat as a mention or a channel-wide notification is shown as plain text.
• Each person can ask a few questions a minute. Past that, it replies (only to you) that it is answering too many questions; wait a minute and ask again.
• Its answer is visible to the channel, so it is told never to quote trace content. Neither your question nor its answer is stored in Neens.
• If it can’t answer, it replies with a short reference such as (ref: 3f2a…); give it to your Neens admin to look up.

Several Neens accounts in one workspace

This only happens when your Neens deployment lets several Neens accounts share one Slack workspace (by default a workspace belongs to one account). If your Slack account is linked in more than one Neens account that shares the same Slack workspace, /neens, link previews and @Neens need to know which one you mean. They ask you to choose once:

/neens use acme

Buttons on cards don’t need this: each card already belongs to one Neens account.

The delivery log

Delivery log at the bottom of the tab lists every message Neens tried to send in the last 30 days, for the channels you can see, with what happened:

When Slack is down, messages wait and are delivered once it recovers; they are not lost.

Disconnect, reconnect and update permissions

• Disconnect (the row’s ⋯ menu) tells you how many channels stop receiving messages, then removes the connection and tells Slack to revoke Neens’ access. Your channels and rules stay in Neens, so reconnecting later brings them back without setting anything up again.
• A Slack workspace can be connected to only one Neens company at a time: Slack gives the Neens app a single access token per workspace, so a second company would see and could post in the first company’s channels. If your deployment’s operator has allowed a workspace to be shared and another Neens account also uses it, Neens removes only your connection and leaves the app in place for them. The confirmation says so.
• To remove Neens from Slack’s side instead, open Slack’s Manage apps page for your workspace, pick Neens and select Remove App. The connection then shows as Disconnected in Neens.
• Only an admin of the scope a workspace was connected from (or of a scope above it) can reconnect a workspace that was disconnected, so an agent admin can’t undo a company admin’s disconnect.
• An org or agent admin can disconnect a workspace they connected only while no channel outside their scope uses it; otherwise a company-wide admin has to.
• A workspace someone removed from Slack’s side shows as Disconnected. One whose access Slack rejected shows Needs attention. Use Reconnect from the ⋯ menu.
• Deleting an agent removes its channels, rules and delivery history, and disconnects a workspace that was connected from that agent and used only by it.

Demo workspaces

A demo environment comes with a workspace labelled Demo: messages are not sent, with example channels and rules. Neens routes insights through those rules for real, and the delivery log shows what would have been posted as Not sent (demo workspace). Nothing is ever sent to Slack from a demo workspace.

The company admin also has a demo Slack account linked, with the morning inbox DM switched on under Settings → Account → Slack. A demo link can’t act and never receives a message, so /neens, previews and @Neens need a real workspace.

Manage it with the API

Everything on the tab is also available under /api/integrations/, with a session token or an admin’s credentials (a project API key can read but not change anything). For example, to create a rule:

curl -s -X POST https://neens.example.com/api/integrations/chat/rules \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Returns agent: critical",
    "scopeType": "project",
    "scopeId": "<agent id>",
    "minSeverity": "critical",
    "mode": "immediate",
    "destinationIds": ["<channel id>"]
  }'

The full reference, with every field and error code, is in the API reference.

Troubleshooting

Related: What Neens shares with Slack · Insights · Alert rules · Members & roles