Skip to main content
A company connects one Slack workspace, then gives an agent its own Slack app, or puts one router app in front of many agents, so people can talk to agents in channels, threads and direct messages. The integration also keeps the Slack directory that agent and workflow access lists draw on, links Slack people to Olyros users for cost attribution, and queues every outbound message so long or interrupted replies arrive intact. Settings live on each agent’s Slack tab (/agents/:agentId/slack) or a router’s (/routers/:agentId/slack), in the sidebar group Settings under Slack Users (/settings/slack-users) and Slack Groups (/settings/slack-groups, /settings/slack-groups/:groupId), and the workspace connection under Settings -> Integrations (/integrations). Who may talk to an agent is set on its Access tab (/agents/:agentId/access), described in Agents.

​
What it does

  • Workspace connection. A company saves Slack app configuration tokens (saveSlackConfigTokens, testSlackConfigTokens); with them Olyros can create apps programmatically. disconnectSlackWorkspace removes the workspace and clears the stored Slack team, so the company can later connect a different workspace.
  • One Slack app per agent or router. POST /api/slack/agents/{agent_id}/create-app builds the app from a manifest and installs it through Slack OAuth (GET /api/slack/oauth/callback); .../reinstall re-applies the manifest, and upgradeSlackAppHomeManifests upgrades pending App Home manifests (all, or one agent). The install link carries a signed state that expires after 600 seconds and is checked before Olyros talks to Slack: the person who started it must still be a company member, and an install into a different Slack team than the connected workspace is refused (“This Slack app was installed into a different Slack workspace than the one connected to this company.”). An expired link shows “This Slack link expired. Open the agent’s Slack page and click Re-authorize.” The AgentSlackConnection holds the app id, bot user, encrypted bot credentials, display name and the channel list (updateAgentSlackConnection). disconnectAgentFromSlack detaches the app; slackCapableAgents lists agents eligible for Slack.
  • Routers. A router is an agent of kind router with its own Slack app; it hands each person to the agent their route names, and replies, App Home buttons, questions and the busy prompt leave through the router’s bot (via_agent_id records the path). One app can therefore front many agents and sandboxes. Routes and defaults are documented in Routers, not here.
  • Channel list. A non-empty channel list limits the channels the bot answers in: a mention in any other channel gets the ephemeral reply “I’m not enabled in this channel.” and is not processed. Direct messages are not limited by the channel list; the agent’s access list decides them. An empty list means every channel the bot is in. Channel ids must be Slack channel ids (C… or G…). The bot answers mentions only: trigger_mode accepts only mentions_only, and all_messages is rejected.
  • Conversations. Slack events (POST /api/slack/events) and interactive actions (POST /api/slack/interactions) are signature-checked and mapped to the company from the verified workspace before any handler runs, acknowledged within Slack’s budget and handled in the background; bot messages are ignored. Replies post as the agent’s own bot (or the router’s) and are threaded. Per agent, setSlackReplyOptions chooses a reply preset (final_replies_only, status_and_replies, verbose) or individual frames (container_notice, session_metadata, thinking, tool_call_summary, tool_call_arguments, tool_call_result, error_detail, invocation_metrics) and a rich-rendering preset (plain, standard, rich; toggles final_reply_tables, final_reply_markdown).
  • Outbound queue. Every message goes through SlackOutboundMessage (pending with retries, attempts and next_retry_at; long replies are split into chunks with chunk_sequence / chunk_total; files attached by id). A platform admin can inspect and requeue (/api/admin/slack-outbound/..., admin app /slack-outbound). A second message sent while the agent is still busy is not dropped: the person gets an ephemeral prompt with Wait or Stop & send, and the pending message moves through awaiting_decision, queued_for_after, dispatched or expired.
  • Sandbox agents from Slack. App Home and message buttons drive sandbox lifecycle actions start, stop, rebuild, open_in_olyros, invocation_stop and rebuild confirmations; each click is logged with result ok, rejected_unauthorized, rejected_state or error.
  • Questions to people. When an agent asks a structured question (olyros_ask_user), it can be delivered as a Slack card; the click resumes the agent. Delivery channels are web_chat, slack, text.
  • Slack users. The workspace user list is cached (SlackUserCache: names, email, avatar, bot flag, last activity) and refreshed per agent (refreshSlackWorkspaceUsers) or registered by email (registerSlackUserByEmail). A cached person is linked to an Olyros user automatically when the Slack profile is trusted (home team of the installation, not a guest, email confirmed) and its email matches a verified Olyros email. Admins browse and export it (slackUsersPage with search, linked filter and tags; slackUsersActivityExport), link a Slack person to an Olyros user (linkSlackUser, unlinkSlackUser), tag them for cost attribution (setSlackUserTags) and delete stale entries.
  • Slack groups and access. Groups (createSlackUserGroup, add and remove members, addTaggedSlackUsersToGroup to build a group from a tag; origin manual or tag, with an optional Slack usergroup id) are subjects of access grants. An agent’s Access tab (setAgentAccess) and a workflow’s Access table can list Olyros users, Slack users and Slack groups; an object with no grant is open to everyone, one with any grant only to the listed subjects, on the web and in Slack alike (see Agents). A Slack user also passes through the Olyros user it is linked to. deleteSlackUserGroup refuses a group that any agent or workflow still grants, naming them. Admins audit who can reach which agent (slackAccessAuditAgents, slackAccessAudit); agentSlackStats shows per-agent Slack usage. The pickers show only people from the currently connected workspace (slackUsersPage(currentWorkspaceOnly)); the Slack Users page shows every cached row.
  • Workflows. The slack_message node posts to a channel or user as a chosen agent; see Workflows.

​
Who can use it

Talking to an agent in Slack is decided per agent by its Access table (ability agent.chat), not by role. The abilities and their roles are listed under Settings -> Roles (/settings/roles). No plan or tier gate.

​
How to configure it

  1. Settings -> Integrations -> Slack: paste the app configuration access and refresh tokens from Slack’s app settings (saveSlackConfigTokens) and test them.
  2. On an agent’s Slack tab (/agents/:agentId/slack), Create app (POST /api/slack/agents/{agent_id}/create-app) and approve the install in Slack within ten minutes; then optionally list the channels it may answer in (updateAgentSlackConnection) and choose the reply and rich presets (setSlackReplyOptions). To serve many agents through one app, create a router under Agents -> Routers (/routers) and do this on the router’s Slack tab instead (Routers).
  3. Invite the bot to the channels you listed.
  4. Settings -> Slack Users (/settings/slack-users): refresh the directory, link people to Olyros users, tag them; Settings -> Slack Groups (/settings/slack-groups): build groups.
  5. On the agent’s Access tab (/agents/:agentId/access, setAgentAccess) list the Olyros users, Slack users or Slack groups who may talk to it, then review the access audit.
  6. After a manifest change, run the App Home upgrade (upgradeSlackAppHomeManifests).

​
Limits and defaults

​
What it emits

  • Agent conversations and tool activity rendered in Slack threads, per the reply options.
  • Outbound message rows (status, attempts, chunks, posted timestamps) and pending message rows; processed-event records for dedup.
  • Lifecycle action logs for sandbox buttons; HITL requests and answers.
  • Slack user and group tables used by access control and cost attribution.

​
Known gaps

  • Everything is per workspace; a company cannot connect two Slack workspaces at once.
  • The bot cannot listen to every message in a channel; there is no all-messages mode and only mentions and DMs reach the agent.
  • Access audit and user management are admin-only and app creation needs an editor, but the workspace tokens and disconnecting a workspace or an agent’s app are open to any member.
  • There is no Slack trigger node for workflows; Slack starts agents, and agents can start workflows through the api_trigger tools.
  • Automatic linking needs a matching verified email; there is no matching by SSO identity, and guests or Slack Connect users are never linked automatically.

​
Customer-safe summary

Connect your Slack workspace and give each agent its own Slack app, or put one shared app in front of many agents. People mention the agent in the channels you choose, or message it directly, and it answers in a thread. You decide how much it shows while working, from final answers only to full progress and tool activity, and how richly it formats results. Long answers arrive in order, and if you message an agent that is still working you can choose to wait or to interrupt it. Admins see who in Slack is using which agent, group people, restrict agents to the right users, and attribute usage to the people and teams behind it.