Use case

A shared brain for your team, on Slack

Connect Notion, Confluence, or Drive. Any teammate can ask the agent how something works and get a cited answer back. The agent can also file the Linear ticket, post in the right channel, or open a PR to fix the doc when it's wrong.

Try Nairi free See Slack integration

What this looks like

Every team has a senior person who answers the same questions over and over. New-joiner onboarding, how the access-request flow works, what the SLA is on enterprise plans, why the deploy pipeline is set up the way it is. The answers are usually in the wiki. Nobody has time to remember which page.

Nairi sits in an ask-channel and answers from your actual wiki. Notion, Confluence, Drive, the codebase, all read through MCP. Every reply cites the source doc. When the docs are wrong, the agent surfaces the contradiction or opens a PR to fix them. When the next step is "file a ticket" or "post in the access-requests channel", the agent does that too, after confirmation.

Confidentiality is built in. Scope the MCP install to specific Notion spaces or Drive folders, add a confidentiality rule covering anything sensitive, and the agent refuses to share what isn't for the room. Multi-person by default, owned by the org, mentionable by anyone.

What the agent does

Reading is half. Acting on what it reads is the other half.

Answer questions from your wiki, with citations

Mention @Nairi in any channel: "how do I get prod access?" or "what is our SLA for enterprise plans?". The agent searches Notion, Confluence, or Drive, and replies with the answer plus a link to the source doc.

Read the codebase too

CLAUDE.md, AGENTS.md, READMEs, /docs folders, internal architecture notes. The agent treats the codebase as another knowledge source for engineering questions, so "how does our auth flow work?" gets a real answer with code citations.

File the ticket on the spot

The agent can act on what it reads. "I need access to the analytics dashboard" becomes a one-line confirmation, a Linear ticket filed in the right team's queue, and a link back in the Slack thread. Same for Jira, GitHub issues, or cross-channel Slack posts.

Open a PR to update the docs when they're wrong

When the docs contradict reality, the agent can open a PR against the wiki repo (for docs-as-code teams) or file a wiki-maintenance ticket. The longer the agent runs, the more the docs improve.

Surface the conflict instead of guessing

If two docs disagree about the same thing, the agent says so and links both. No quiet hallucination, no picking one source and pretending the other doesn't exist.

Confidentiality dial built in

Rules tell the agent what not to share. Customer PII, hiring data, comp bands, board materials, anything you want kept inside specific channels. The agent refuses and points to the right human.

Three scenarios teams use it for

From the first day to the docs-maintenance loop.

New joiner, day one

Drop the new hire into a private channel with the agent on day one. They get a 24/7 buddy for their first two weeks who knows the onboarding checklist, the access-request process, the team Slack channels, and where to find the architecture docs. The team doesn't spend the first week answering the same five questions for every new person.

Cross-team translation

Sales asks in #ask-anything: "what is our SLA on enterprise plans?". The agent reads the sales playbook and the engineering on-call doc, returns the canonical answer with both citations. No-one had to context-switch into Notion, find the page, paste the link.

The docs are wrong

Engineer asks how the auth refresh flow works; agent answers from the wiki, engineer says "that's out of date, here's how it actually works now". Agent opens a PR against the wiki repo with the correction (for docs-as-code teams) or files a wiki-maintenance ticket. The knowledge base gets sharper every week.

How it works

One agent, scoped to your knowledge sources and your local policy.

1
Connect the wiki via MCP
Notion, Confluence, or Google Drive, pick the one your team actually writes in. Most wikis are one-click OAuth via the MCP marketplace. Scope tight: pick the specific spaces or folders the agent is allowed to read. OAuth credentials sit at the org level; every agent can use the same connection.
2
Connect the action tools
Linear, Jira, GitHub, Slack writes. Pick the two verbs your team uses most ("file a ticket", "post in a channel"). Same one-click marketplace install. Reading is half the value; the other half is the agent doing the next step without a human handoff.
3
Encode the local policy as rules
Knowledge map (which source covers which topic), confidentiality (what not to share), escalation (who owns what), action safety (confirm before write actions, never delete or archive). The base prompt is identity; the rules are how your team actually works.
4
Bind it to an ask-channel and let the team use it
#ask-nairi, #ask-anything, #onboarding, your call. The agent reads the conversation context, replies in the thread, asks for confirmation before any write action. Multi-person by default. Everyone in the channel can ask the agent things.

Put a shared brain in your ask-channel

Connect the wiki, encode the local policy, mention the agent. Cited answers, confirmation before write actions, self-host the runtime if your wiki can't leave your infra.

Try Nairi free View pricing

Questions about company-knowledge use

What teams ask before they wire up the wiki.

The senior person who knows everything and keeps getting interrupted to answer the same questions. Not in a hostile way; they're still the source of truth for the things the docs don't cover. The agent just handles the questions that are already documented (or should be), so the human gets pulled in for the actual judgment calls, not for "where is the link to the onboarding doc again?". Most teams see the bulk of new-joiner questions, access requests, and cross-team policy questions move to the agent within a few weeks.

Notion, Confluence, Google Drive, and any GitHub repo (for CLAUDE.md, AGENTS.md, READMEs, /docs folders). Most wikis are one-click OAuth via the MCP marketplace. If your team's wiki is something niche or internal, it can usually be wired up as a custom MCP with an API token. Linear and Jira work too if you want the agent to read tickets as additional context.

Both. Connect Linear or Jira from the MCP marketplace and the agent can file tickets after the user confirms in the same thread. Connect the Slack actions MCP and it can post in other channels. Connect the GitHub MCP and it can open PRs against the wiki repo (for docs-as-code teams). Write actions are always behind a confirmation step by default; the agent asks "file this in Linear's product-platform team, ok?" and waits for "yes" before doing it.

Two layers. Scope at install time: when you connect Notion or Drive, pick only the spaces/folders the agent is allowed to read. The rest stays invisible to the agent. Plus a confidentiality rule at the agent level: "never paste customer PII, hiring pipeline data, comp data outside published bands, anything from the Board materials space". The agent refuses and points to the right human. Both layers stack: the MCP can't see what isn't scoped in, and the agent won't share what the rule forbids even if it could see it.

Two structural answers. First, every reply cites the source doc, so the user can verify in one click. Second, the base prompt explicitly tells the agent: if you can't find the answer in the wiki, say so and recommend who to ask. Do not invent policy. If two docs contradict, surface the conflict instead of picking one. This is a behavior question more than a model question, and the rules carry the discipline.

Glean and Notion AI are dedicated knowledge-search products and they're good at indexing-heavy retrieval across many sources. Where Nairi fits differently: it's a general team agent that also does knowledge Q&A. The same agent that answers "how do we do X here?" can also file the Linear ticket, post in the announcements channel, open a PR to fix the doc, run a scheduled job. One platform, one set of credentials, one vault, per-agent customization via skills and rules. If knowledge search is the only thing you need, evaluate the specialists.

Per agent, your choice: Claude (Sonnet 4.6 or Opus 4.7) via Claude Code, GPT-5 via Codex, open-weights via OpenCode, or the Cursor harness. For long-context Q&A across a large wiki, Claude Sonnet is what we recommend by default because the context window and the retrieval behavior fit. Switch if your team has a different preference.

Yes. The nairid daemon is open-source Go and runs the full agent loop, including the sandbox where the wiki MCP runs, on your hardware. Slack messages route through the Nairi backend for delivery, but the actual wiki content and the agent reasoning stay inside the daemon. Useful for regulated industries, for orgs whose wikis contain PII or IP, or for any compliance posture that requires the data to never leave the corporate network.

Connect one wiki (the one with the most-asked questions, usually onboarding + access + policies) and one action tool (Linear or Slack writes). Bind the agent to one channel (#ask-nairi works well). Add a knowledge map rule pointing at the right Notion spaces and a confidentiality rule covering anything sensitive. Use it for two weeks. Then expand: add more sources, add Linear or Jira if you didn't start there, add the docs-PR loop. Doing it incrementally calibrates the rules without overwhelming anyone.

Data analyst on Slack On-call incident response Automated PR review Slack integration Discord integration GitHub integration Pricing Documentation