Collaboration mode

Last9 AI is multiplayer in Slack: several engineers ask questions in the same thread, and the assistant answers all of them in that one investigation. An AI Assistant conversation in the dashboard is your own, and you can share it with a colleague in your Last9 organization to read.

Use a Slack thread when teammates need to ask questions together. Share an AI Assistant conversation when a colleague only needs to read the findings from your dashboard investigation.

Open your conversation in AI Assistant and wait for the response to finish.
Click Share to create and copy a share link.
Send the link to a colleague in the same Last9 organization. Use Copy link to copy it again later.
To revoke access through the link, click Unshare, then confirm with Unshare in the dialog.

The shared view is read-only. A colleague can read the conversation, but cannot send messages or edit your chat through the link. The link opens the saved conversation, so later saved replies can appear when it is opened again.

Only the conversation owner can create or revoke its share link. Recipients must belong to the same Last9 organization, and the link does not provide public access.

Investigate together in Slack

For installation, authorization, channel access, and troubleshooting, see Use Last9 AI in Slack. This section covers what changes once a team shares the thread.

Keep every question about one incident in the same thread, so each engineer gets answers with the same incident context. An engineer can start with the affected service and time window:

@Last9 Investigate HTTP 5xx errors for checkout-api in production
between 14:00 and 14:30 UTC on 2026-09-10. Identify the affected
routes and include links to the supporting logs and traces.

A teammate can then add information or ask a follow-up in the same thread:

@Last9 A deployment completed at 14:05 UTC. Check whether the
failure pattern changed after it. Keep the same service and
incident window, and explain what the evidence supports.

The assistant answers both engineers in that one investigation. Each teammate who queries Last9 through the app needs their own authorized Last9 account, with an email address matching their Slack account. A teammate without one can still read the thread.

Everyone reads the same answers, so state a fresh query window when you check recovery. An earlier answer in the thread describes the data queried at that time, not current health.

For evidence review and recovery checks, refer to Investigate an incident with AI.

Use organization context

Last9 AI uses your organization’s configured skills, access to its telemetry, and tool permissions to investigate requests. It distinguishes reusable memory from conversation history: it reuses durable organization knowledge where that knowledge is configured, and keeps conversation history with the individual chat or incident thread.

An investigation message does not automatically become durable organization knowledge. Sharing a chat also does not combine its history with other dashboard conversations or Slack threads.

Add runbooks and past incident knowledge

Your team can supply reviewed references for organization-specific skills to use during investigations. Useful references include service relationships, environment and region mappings, known error patterns, runbooks, and past root cause analyses (RCAs).

When configured, a skill reads the relevant reference during an investigation. Supplying a reference does not mean its full contents are included in every conversation.

Prepare the references your team wants Last9 AI to use. Keep confirmed findings separate from possible causes, and include supporting evidence.
Contact support@last9.io to configure the references and investigation skills for your organization. For an agent you manage yourself, refer to Build your own AI SRE on Last9 data.
Test a question about a known incident. Ask the agent to identify the reference it used and check that the response matches the reviewed record.
Assign an owner to update each reference when services, runbooks, or findings change. After an update, repeat the test to confirm the agent can read the revised information.

For a past RCA, a short record can contain the following details. The format is illustrative and is not a Last9 API schema:

# Incident title

- Incident window: <start and end time in UTC>
- Affected service and environment: <service, environment, and region>
- Symptoms: <observed failures and user impact>
- Confirmed cause: <cause supported by the investigation>
- Evidence: <queries, trace links, or reviewed incident report>
- Action and outcome: <what changed and how recovery was verified>
- Follow-up: <remaining work or changes to the runbook>
- Owner and review date: <responsible team and last review date>

Use a past RCA to guide what to check, then compare it with current telemetry. A similar error message does not establish the same cause. An investigation should identify the matching evidence, explain any differences, and report what remains unverified.

Manage organization access

An organization admin controls AI access in AI Settings. Anyone with access to a Slack channel reads the answers posted there, so see Channel access and shared information before you choose an incident channel.

Sharing findings does not approve a proposed change. Review any approval request in the interface where the action is proposed, and keep completed actions separate from recommendations in the incident report.

Build a shared agent with Mithai

Mithai is Last9’s open-source multiplayer agent harness for your organization. It powers Last9 AI, and it can also run an agent you build for your own workflows. Mithai keeps conversation history in sessions and reusable notes in memory. Its Slack adapter uses a separate session for each thread. See Mithai core concepts for the session and memory model.

For a custom agent, decide which team facts belong in shared notes and which incident details should stay in the conversation. Configure the skills, tool permissions, model, and adapters the team needs. Mithai also supports multiple independent agents with separate skill sets and memory through its multi-agent configuration.

Preserve the agent’s memory and session storage across restarts, and review who can access both. The deployment guide explains persistent storage, and the security guide covers credentials, tool approvals, and channel access.

To connect a custom agent to Last9 telemetry, refer to Build your own AI SRE on Last9 data.

Troubleshooting

A colleague cannot open the share link

Confirm that the person belongs to the same Last9 organization. If you revoked the link with Unshare, click Share to create a new one.
Share is unavailable in your conversation

Wait for the response to finish. Only the conversation owner can create or revoke its share link.
@Last9 does not reply in the incident channel

Invite the app to the channel and mention @Last9 in each follow-up. See Slack troubleshooting for the workspace connection.
A teammate’s question is not answered

Check that their Slack email address matches their Last9 account email. Ask an organization admin to add the account.
An investigation does not use your runbook or past RCA

Ask the agent to identify the reference it used. Contact support@last9.io to configure the references and investigation skills for your organization.

Please get in touch with us on Discord or Email if you have any questions.