Your company already wrote the runbook.Brain compiles it.

Brain reads the Slack threads, pull requests and tickets where your team’s real procedures live — and distills them into provenance-backed Skills your AI agents can run. Any agent. Any model. Served over MCP.

Join the waitlist See the pipelineOpen standards: MCP · Agent Skills (SKILL.md)
Fig. 01Fragments in · procedure outIllustrative
#incidents — Dec 9PR #241 reviewOPS-118 ticketthread, 212 repliespostmortem draftwiki, stalereview noteDM, lostmacroBRAIN9-STAGE DISTILLERRECURRENCE FOUNDCONFLICTS RESOLVED · NOISE DROPPEDSKILL.mdPROVENANCE: 31 EVENTS
Sources: Slack · GitHub · Jira · Gmail · Zendesk · ClickUp · BitbucketServed over MCP → any agent
Distilled (sample)

7

Source connectors at launch

5

Extractors working every cluster in parallel

9

Pipeline stages, raw event to skill

0

Agents locked in — any MCP client


01The problem

The knowledge exists. It just sank.

The procedures that actually run your company were never written down as procedures. They happened — in threads, reviews and tickets — and then the scroll swallowed them.

Slack · six months ago

The OOM fix is in a thread from December.

Someone solved it, someone thanked them, and it sank. The next incident starts from zero.

Tribal knowledge

The deploy checklist lives in one engineer’s head.

The real procedure is whatever the last person who ran it remembers. When they’re on holiday, so is the process.

Agents today

Your agent confidently invents a process.

Agents are trained on the internet, not on your company. Without your history they guess — fluently, plausibly, wrong.

Onboarding

Every new hire re-derives everything.

Months of asking around to reconstruct knowledge the organization already paid to learn once.


02How it works

Connect. Distill. Serve.

Read-only ingestion in, portable skills out. Click a stage — the instrument follows.

Fig. 02The pipelineLive schematic
SlackGitHubJira+ 4 moreREAD-ONLY · PULLS + WEBHOOKS + BACKFILLDISTILL5 EXTRACTORS · IN PARALLELRECURRENCE · CONFLICTS · NOISESKILL.md+ SOURCE EVENT IDSMCPPER-TENANTClaude CodeCursorIn-house
1Normalize2Enrich3Segment4Classify5Embed6Cluster7Extract8Resolve9Compile

03The moment

Same agent. Same question. One has your history.

Without BrainIllustrative
You

payments-service is OOMing in prod — what do I do?

Agent

Here are some general steps you can try:

  1. Check the application logs for errors.
  2. Restart the affected pods.
  3. Consider increasing the memory limits.
  4. Review recent deployments for changes.
  5. Monitor the service after restarting.

If the issue persists, consult your team’s documentation.

With BrainIllustrative
You

payments-service is OOMing in prod — what do I do?

Agent · loaded skill: payments-oom-runbook

Running your team’s runbook for this exact failure:

  1. Check the processor status page first — upstream degradation causes the retry pile-up.
  2. Drain the retry queue: payments-cli queue drain --older-than 1h
  3. Scale to 6 pods while the queue clears.
  4. If memory isn’t stable in 15 min, escalate to the payments on-call lead.
Provenance— distilled from 31 events: #incidents Dec 9 + Feb 17, PR #241
Both transcripts are mock-ups of the same agent, with and without Brain’s MCP server connected.

04Provenance

Nothing ships without its sources.

Every Skill carries the event IDs it was distilled from, a confidence score and an extraction timestamp — so a reviewer, an auditor, or the agent itself can walk any instruction back to the conversation it came from.

  1. 1Check the processor status page first.
  2. 2Drain the retry queue, oldest hour first.
  3. 3Scale to 6 pods while the queue clears.
  4. 4Not stable in 15 min → escalate to on-call lead.

Hover a step — its source events light up.

skills/payments-oom-runbook/SKILL.mdIllustrative
---name: payments-oom-runbookdescription: Use when payments-service is OOMing or restarting under memory pressure in prod.source_event_ids: - slack:incidents:2025-12-09T14:02 - slack:incidents:2026-02-17T09:41 - github:pr:241 # 28 more eventsconfidence: 0.85last_extracted: 2026-06-28extractor_version: brain/0.3--- # Payments OOM runbook 1. Check the processor status page first

05Connectors

Plugged into where work happens.

Seven sources at launch, more behind one interface: a connector is a single contract — list resources, fetch events. Adding yours is an afternoon, not an integration project.

SlackChat · threadsLive
GitHubCode · PRs · reviewsLive
JiraTickets · sprintsLive
GmailEmailLive
ZendeskSupport ticketsLive
ClickUpProjects · tasksLive
BitbucketCode · PRsLive
LinearTicketsNext
NotionDocs · wikisNext
IntercomSupport chatNext
PagerDutyIncidentsNext
YoursOne interface · ~200 linesBrainConnector

06Who it’s for

Three front doors. One brain.

On-call that remembers every incident you ever resolved.

payments-oom-runbook

The fix that worked, as a procedure: drain the retry queue, scale, escalate — with the thresholds your team actually used.

31 events

deploy-checklist

The real order of operations, including the migration gate nobody ever wrote down.

24 events

code-review-norms

What your senior reviewers actually enforce, distilled from a year of review comments.

41 events

ConnectorsGitHub · Jira · Slack · PagerDuty


Glean helps your people find answers.Brain teaches your agents how your company works.

Where Brain sits

Give your agents a memory.

Early tenants get white-glove connector setup and a distilled skill library from their first six months of history.