Self-hosted · Read-only by design · Zero data egress

The memory layer
your organization never had.

Every organization runs on knowledge that never gets written down — the decision made in a meeting, the fix found at 2am, the workaround only one person knows. It lives in conversations, tickets, and people's heads, and it walks out the door when they do. DocBrain captures that knowledge where it's created, connects it across the tools you already use, and keeps it accurate. So expertise survives turnover, new people ramp in weeks instead of months, and your teams stop re-solving problems they already solved.

13+ sources, one memory Zero data egress Read-only by design Cited answers, always
Scroll
0+
Knowledge sources, unified
Zero
Data leaves your network
Read-only
Access to your systems
Cited
Every answer, to its source

The hidden cost

The knowledge you're losing
right now.

It isn't a documentation problem. The knowledge that runs your organization was never written down in the first place — it lives in conversations, tickets, and the heads of your most experienced people. Every day, some of it disappears.

Your experts leave, and take years with them

The engineer who designed the system, the operator who knows the workaround, the lead who remembers why. When they go, decades of hard-won context walk out with them — and no wiki ever captured it.

New people take months to get up to speed

Onboarding runs on interruptions. Every new hire re-asks questions someone already answered, because the answers were never captured anywhere they can find them.

The same problems get solved over and over

A team hits an issue another team fixed last quarter, and burns days rediscovering the answer — because the knowledge was fragmented across tools nobody thinks to search.

And it's getting more expensive. People change roles faster, teams work across time zones, and AI now produces work faster than any human can absorb the reasoning behind it. The gap between what your organization does and what it remembers is widening every quarter.

Why the old way fails

Why the tools you have
don't fix this.

You've already bought places to store knowledge. The problem is what they can capture — and what they can't.

Wikis & knowledge bases

Written by hand, after the fact

  • × Only holds what someone remembered to write down
  • × Stale the moment the work moves on
  • × Disconnected from where knowledge is actually created
  • × No idea which pages are still true
A snapshot that decays.
AI chatbots & search

A bot bolted onto the same wiki

  • × Answers only from what's already written
  • × Retrieves faster, but the source is still stale
  • × Forgets every question the moment it's answered
  • × Confidently makes things up when the doc is missing
Retrieves a stale wiki faster.
DocBrain

Captured where knowledge is created

  • Captures the knowledge that was never written down
  • Connects it across every tool into one memory
  • Tracks freshness and flags what's gone stale
  • Preserves the decisions and the why
Memory that compounds.

Existing tools index what was written down. The knowledge that actually runs your organization never was. That's the layer DocBrain builds.

How It Works

How DocBrain builds
your organization's memory

Knowledge captured where it's created, connected into one memory, preserved with its context, and kept accurate over time — capturing knowledge at the source instead of documenting after the fact.

The difference
01

Capture

Knowledge is captured the moment it's created — from the conversations, tickets, incidents, and work systems your teams already use. Nothing to write. No wiki to maintain.

Knowledge never written down
02

Connect

Every fragment is linked into a single connected memory — by topic, by who knows it, by what depends on what. Not a pile of documents; a graph you can ask.

One connected memory
03

Preserve

Decisions, fixes, and the reasoning behind them are kept with their provenance — so the why survives long after the person who knew it has moved on.

The why, kept
04

Keep current

Freshness is tracked continuously. When something goes stale or two sources contradict, DocBrain flags it — so you can trust what the memory tells you.

Always know what's true
What no one else does

Memory, not another archive

Other tools organize the knowledge you already wrote down. DocBrain captures the knowledge you didn't — the decisions, fixes, and expertise that only ever lived in a conversation or someone's head — and turns it into memory your whole organization can draw on.

Result: a living memory grounded in real work, kept current, without asking anyone to stop and write docs.

On demand · docbrain generate

Turn your memory into
a document, on demand.

docbrain generate drafts a runbook, a postmortem, an onboarding guide, or a reference — grounded in your organization's own memory: your incidents, tickets, decisions, and past work, with per-claim provenance. A general AI model writes fluent, generic prose. generate writes what is true for you, or tells you honestly what it can't answer. Run it from your terminal, or wire it into your pipeline so documents update themselves as work happens.

Local — draft a runbook from your notes
docbrain-cli
$ docbrain generate "runbook for cert rotation" --source notes.md --type runbook > runbook.md
⌛ Gathering org context (corpus + episodes)…
⌛ Drafting · resolving provenance per section…
doc_type: runbook quality: 87/100 (0 errors)
provenance: Overview ← Deploy Runbook v3 · Steps ← Cert Rotation SOP
needs_input: "Which CA issues the production certs?"
runbook.md written · 142 lines, grounded, ready to review

stdout is the markdown (pipe-clean). Diagnostics go to stderr. Swap --source notes.md for --source-url to ground in a Confluence page, Jira issue, Slack thread, or GitHub PR.

CI — update the runbook on every merge
.github/workflows/docs-on-merge.yml
- name: Update runbook from the merged PR
run: |
docbrain generate "update the deploy runbook for this PR" \
--source-url https://github.com/acme/payments/pull/$PR \
--target https://acme.atlassian.net/…/Deploy+Runbook \
--type runbook --out runbook-draft.md
⌛ Fetching PR via the GitHub connector…
⌛ Augmenting the existing runbook (not overwriting)…
runbook-draft.md · posted to the PR for review

Drop --out and the exit code becomes a gate: generate exits non-zero on error-severity quality violations, so a sub-bar doc fails the build. If a named source can't be fetched, the whole run hard-fails — never a doc from a partial set.

Grounded, with provenance

Reads your real PRs, runbooks, incidents, and threads through your connectors. Every section is attributed to the source it came from — reviewers verify, they don't trust.

Honest when it doesn't know

When the knowledge isn't there, it emits needs_input — the open questions — instead of fabricating. You fix gaps on purpose, not mid-incident.

Same gates, in CI

Secret/PII redaction, hostname scrub, injection-quarantine, and a 0–100 quality score run on every draft. The exit code is your CI signal — "good enough to publish" becomes machine-checkable.

Access Everywhere

Meets you where
you work

One memory, many ways in. Ask from chat during an incident, from your editor while you work, from the terminal, or from the web dashboard — everyone drawing on the same shared memory.

docbrain-cli
# engineering
JC
Jake Chen2:14 PM
Payments 502 after the deploy. Has anyone seen this before?
DB
DocBrain2:14 PM
Runbook: Payments Service Rollback Procedure
1. Check deploy diff: git log --oneline production..main
2. Rollback: kubectl rollout undo deployment/payment-processor
3. Verify: curl -sf https://api.example.com/health/payments
Sources: Payments On-Call Guide, Deploy Runbook v3 · Confidence: 91%
👍 4🏆 2
feat: add idempotency keys to refund endpoint Merged
DB
DocBrain Bot commented 2 minutes ago
3 knowledge fragments extracted
1. Decision: Idempotency keys prevent duplicate refund processing
2. Caveat: Keys expire after 24h; retry must generate new key
3. Procedure: Manual refund override via admin API requires manager approval
Confidence: 92% · Auto-indexed to "Payments" space
payment_service.rs
refund_handler.rs
pub async fn process_refund(req: RefundRequest) {
  // Idempotency check
  if let Some(existing) = cache.get(&req.idempotency_key) {
    return Ok(existing.clone());
  }
  // Process refund...
}

Coding agents · Claude Code & Cursor

Your agent fixes it.
DocBrain remembers it.

Engineers now solve their hardest problems inside agent sessions — and that knowledge dies when the terminal closes. DocBrain's MCP tools give your agent a write path: it checks whether the org already knows the fix, drafts a capture, and asks you before filing it. Three lines in your CLAUDE.md:

CLAUDE.md
When we resolve an error or discover non-obvious behavior, call
docbrain_suggest_capture for the files involved. If a gap exists,
draft a 3–5 line capture and ask me to approve it before calling
docbrain_capture. Never include secrets or .env values in a capture.

Nothing uploads automatically — the agent proposes, you approve, and the capture lands in the normal review queue. The client is MIT-licensed: read the full guide.

Live Intelligence

Answers from
what's true right now

Most knowledge tools answer from a snapshot. An index tells you what was true the last time it crawled, not what's happening this minute.

DocBrain reaches into your live systems at the moment you ask. “What's the status of PROJ‑412?” pulls the ticket from Jira as it stands right now, then combines it with your indexed runbooks into one cited answer. Old knowledge and live state, in a single response.

Read-only by design

DocBrain can read your tools, never write to them. Any tool that could change a system is dropped at discovery. Enforced, not promised.

Each user, their own permissions

Connected per‑user over OAuth. Your answer is scoped to what you can see in the source. Tokens refresh on their own, and nobody sees data they shouldn't.

Any MCP-enabled system

Jira, Confluence, and Slack out of the box, plus anything that speaks the Model Context Protocol. Add a tool, and DocBrain starts answering from it.

0

Write operations

Per‑user

OAuth scope

Live

At answer time

DB
DocBrain Live data

“Is the payments outage from yesterday resolved, and what was the fix?”

Resolved, 14h ago.

PROJ‑412 (“Payments 502s after deploy”) moved to Done at 09:14 UTC. The fix was a rollback of payment-processor plus an idempotency-key guard, per the linked runbook.

Jira · PROJ-412 live Payments Rollback Runbook v3 PR #347
Indexed knowledge
The runbook & PR. What was true at capture.
Live tool
Ticket status right now. Pulled at answer time.

One answer. Indexed history reconciled with live source-system state.

A question answered vs. a memory that grows

Any AI assistant can look something up for you. The difference is what happens after the answer. It forgets. DocBrain keeps it, connects it, and gets better every time someone asks.

AI assistant
DocBrain
Reads live tools at answer time
Yes
Yes
Remembers the answer after you close the tab
No
Yes
Learns from everyone, not just you
No
Yes
Maps who owns what, what depends on what
No
Yes
Turns answers into durable, quality-scored docs
No
Yes
One shared memory for the organization, with access control
Per-user
Org-wide

An assistant asks your tools a question. DocBrain turns every question your organization has ever asked into a memory that gets smarter.

4-Layer Memory System

A brain that
remembers everything

Four memory layers work together so every question makes the system sharper. The 100th question about a topic is faster and better-answered than the first.

~40%

Cache hit rate

<3s

Avg response

25MB

Binary (Rust)

Working Memory Redis

Multi-turn conversation context. Resolves "it", "that service", "the same thing."

Episodic Memory Postgres + OpenSearch

Every Q&A ever asked. Semantic caching, feedback learning. Gets cheaper as it learns.

Semantic Memory Knowledge Graph

Entity relationships via graph traversal. "Who owns the payment service?" gets resolved, not searched.

Procedural Memory Learned Rules

Adapts retrieval from feedback. Discovers: "deployment questions → search DevOps first."

Capabilities

Everything the memory layer does

Knowledge Fragments

First-class knowledge units with provenance. Auto-captured from PRs, conversations, and CI pipelines.

CI/CD Capture

Merged PRs and deployments auto-analyzed by LLM. Decisions, caveats, procedures extracted as fragments.

Quality Pipeline

Three-layer scoring: structural, style rules, and LLM-assessed semantic quality. Composite score 0-100.

Confidence-Scored Q&A

Hybrid search, intent classification, cited answers with freshness. Low confidence asks questions instead of hallucinating.

Knowledge Graph

Entity relationships with BFS/DFS traversal. Blast radius analysis, expertise routing.

Documentation Autopilot

Clusters unanswered questions, detects gaps, auto-drafts missing documentation.

Generate Grounded Docs On Demand

docbrain generate drafts a runbook, postmortem, or reference grounded in your own runbooks, incidents, tickets, and PRs — with per-claim provenance, or an honest list of what it can't answer instead of confident fiction. Same redaction, injection-quarantine, and quality gates as every DocBrain doc. Returns the markdown — you decide what to publish. Wire it into CI to update docs on merge.

14 LLM Providers

Anthropic, OpenAI, Bedrock, Ollama (local), Gemini, Azure, and 8 more. Swap via config.

SSO & RBAC

GitHub/GitLab/OIDC SSO. Four roles. Space-level isolation. API keys Argon2-hashed.

Source-System ACL

Mirrors real Confluence / Slack / GitHub / Jira permissions at query time. Restricted page in source = filtered out for users who can't read it. Three modes, side-channel-safe, audit-logged.

Built with Rust

25MB binary. <500ms cold start. Under 100MB memory. Self-hosted, your data stays yours.

Connector SDK

Plug in any knowledge source in any language. 3 endpoints, DocBrain handles the rest.

MCP IDE Capture

10 tools for Claude Code, Cursor, and any MCP editor. Capture decisions at commit time.

Live MCP Tool Platform

Pull live Jira tickets, GitHub PR state, or any MCP-enabled system at answer time. Per-user OAuth with transparent refresh. Answers cite live data, not stale chunks. See how →

13+ Knowledge Sources

Confluence, Slack, Teams, GitHub, GitLab, Jira, PagerDuty, OpsGenie, Zendesk, and more.

Security & enterprise readiness

Built to pass
your security review

DocBrain is designed for the teams whose job is to say no. It runs entirely inside your infrastructure, reads your systems without ever writing to them, and shows its work on every answer.

Self-hosted, zero egress

Runs entirely in your infrastructure from Docker images and Helm charts you pin and control. Point it at a local model and nothing ever leaves your network. Air-gap capable.

Read-only by design

When DocBrain connects to a live system, any tool that could create, edit, or delete is dropped at discovery. It can read to answer a question. It cannot change anything. Enforced, not promised.

Access that mirrors your sources

Source-system ACL mirrors the real permissions of Confluence, Slack, GitHub, and Jira at query time; per-user OAuth scopes live results to what each person can already see. Side-channel-safe, audit-logged.

Identity & governance

SSO via GitHub, GitLab, or any OIDC provider. Four-tier RBAC, space isolation between teams, API keys hashed with Argon2, and every privileged action audit-logged.

Honest by architecture

Every answer is cited to its source. When the knowledge isn't there, DocBrain asks a question instead of inventing one. The design goal is "roughly 0% confidently wrong," not "always right."

Published threat model

Eleven analyzed attack vectors and an operator security checklist, open today. Because DocBrain runs inside your boundary, you can audit the boundary — not just the codebase.

Deployment model

Deploy with Docker Compose or Helm. PostgreSQL, OpenSearch, and Redis run inside your network; the AI model runs wherever you choose — a local model for zero egress, your own cloud account (Bedrock, Azure, Vertex) for your KMS and audit trail, or a provider API for the fastest setup. You decide where your data goes, source by source.

The client source — the CLI and the IDE MCP server — is open in the repo (MIT), so you can audit what runs in your environment and what leaves it. The server source is not published yet — we missed our original H1 2026 target, and we won't post a new date until we're certain of it. The rest of what determines your risk is open too: the deployment you control, the threat model, and the read-only, zero-egress boundary. How DocBrain earns trust →

Use cases

Where a memory layer
pays off

Anywhere critical knowledge lives in people's heads and conversations instead of a system you can search — engineering, operations, support, and beyond.

Onboarding

New people answer their own questions from day one, drawing on everything the team already figured out. Ramp in weeks, not months — without pulling a senior person off their work.

Incident & outage response

During an incident, ask what broke last time and what fixed it. Answers combine past incidents with live system state, cited — so the fix doesn't depend on who happens to be online.

Preserving decisions

Every "why did we do it this way?" has an answer: the decision, the context that shaped it, and who made it — kept with provenance instead of lost in a two-year-old thread.

Offboarding & continuity

Capture what your most experienced people know before they leave. When they do, the memory stays — and the next person inherits their context, not a blank page.

Support & operations

Front-line teams resolve issues from the whole organization's history instead of escalating the same question twice. The answer someone found once is available to everyone after.

Cross-team knowledge

Stop re-solving what another team already solved. One connected memory spans every tool and team, so a fix found in one corner of the organization is findable from every other.

In practice

A day with DocBrain

One worked example, from an engineering team. Knowledge captured where the work happens — not after, not in a doc sprint. At the source, as it happens.

1

9:15 AM · PR Merged

Engineer merges a payment refund PR

DocBrain's CI capture extracts 3 knowledge fragments: a decision about retry logic, a caveat about idempotency keys, and a procedure for manual refund overrides. All auto-indexed with 92% confidence.

2

11:30 AM · Slack Thread

Support asks "how do refunds work now?"

DocBrain answers with fragments captured 2 hours ago, plus links to the original PR. The support engineer types /docbrain capture to save more context.

3

2:00 PM · Auto-Composition

5 fragments cluster into a draft doc

DocBrain detects semantic similarity across PR fragments, Slack capture, and older fragments. Composes "Payment Refund Procedures," scores at 78/100, routes for SME review.

4

3:30 PM · Published

Reviewed, style-checked, published

Quality rules catch 2 violations, auto-fixed. Final score: 91/100. Published to "Payments" space. Total engineer effort: zero.

The old way: schedule a doc sprint, assign writers, review in 2 weeks, stale in 2 months.

With DocBrain: captured at 9:15 AM, part of your memory by 3:30 PM. Kept accurate from then on.

Recorded, not staged

The quickstart, unedited — 90 seconds

Install → ingest → a cited answer → generate turning raw on-call notes into a runbook that cites its sources. Shipped images, 100% local models, sped up only where the title cards say so.

Independent Assessment

We asked Rovo to compare itself.
Honestly.

No cherry-picked benchmarks. We opened Rovo, asked it to do a fair comparison, and published the full response — including where it wins.

"

DocBrain's 'capture knowledge that was never written down' is solving a problem I fundamentally can't.

— Rovo (Atlassian's AI assistant), responding to a direct comparison prompt

Where DocBrain wins
Where Rovo wins
Captures knowledge that was never written down
Native Atlassian integration (Jira, Confluence, Bitbucket)
Doc quality scoring (0–100 pipeline)
Zero setup for existing Atlassian Cloud teams
Self-hosted — your data never leaves your infra
Broad work execution (sprint planning, triage, reviews)
14 LLM providers, including local Ollama for air-gap
Cross-tool action orchestration
Auto-composes docs from fragments — without being asked
3M+ users, Fortune 500 enterprise reach

Full response published unedited. Rovo's assessment, not ours. Read the full transcript →

FAQ

Common questions

What exactly is DocBrain?
An institutional memory layer for your organization. It captures the knowledge your teams create as they work — decisions, fixes, answers, context — from the tools you already use, connects it into one searchable memory, keeps it accurate, and preserves it when people leave. It sits on top of your existing systems, reads them (never writes), and doesn't ask anyone to move to a new tool.
We already have Confluence, Notion, GitHub, and Slack. Why do we need this?
Those are places to store knowledge someone remembered to write down. DocBrain captures what never gets stored — the reasoning in a ticket, the fix in an incident thread, the decision made in a meeting — connects it across all of them, keeps it fresh, and tells you when it's gone stale. You're not replacing Confluence; you're capturing the knowledge Confluence never sees. DocBrain reads your existing tools; it doesn't ask your teams to adopt another one.
Is DocBrain really free?
Yes, free for production use under BSL 1.1. Only restriction: you cannot offer DocBrain as a hosted service. The source code publishes in H1 2026; today you get versioned Docker images, Helm charts, and full docs to self-host. The license converts to Apache 2.0 at 5k stars or Jan 1, 2028.
How do I know the data is accurate and not made up?
DocBrain is built to refuse to show anything it can't stand behind. Confident ownership only shows up after it's been measured against a human-checked answer key and passed a strict accuracy bar. Until then you see "no confident owner," not a guess. Every confident claim carries its evidence. Low-confidence answers ask a question instead of making something up, and the system never passes off intent as a measured result. The honest promise is "roughly 0% confidently wrong," not "always right." Read how DocBrain earns trust →
Does my data leave my infrastructure?
When self-hosted, docs and queries never leave your infra. Use Ollama for 100% local, air-gapped deployment with zero data egress.
Which LLMs does DocBrain support?
14 providers: Anthropic (Claude), OpenAI, AWS Bedrock, Vertex AI, Azure OpenAI, Groq, OpenRouter, Together, DeepSeek, Mistral, xAI, Gemini, Cohere, and Ollama. Swap via config.
Does DocBrain learn from feedback?
Yes. Thumbs up/down updates procedural retrieval rules and feeds Autopilot gap detection. Optional learning pipeline fine-tunes embeddings from feedback.
How long does setup take?
Docker Compose: about 5 minutes. Run bash scripts/setup.sh and it brings up PostgreSQL, OpenSearch, Redis, migrations, and sample docs. Add your LLM API key and go.
What is "capturing knowledge at the source"?
It's the mechanism behind the memory layer — the same idea as shift-left testing. Capture knowledge the moment it's created, where it's created, instead of documenting after the fact. Fragments accumulate, connect, and compose into memory, without asking anyone to stop and write docs.
My IDE already has MCP. Why DocBrain?
Your IDE can read live tools, and so can DocBrain. The difference is memory. Cursor or Claude Code answers your question in your session, then forgets it the moment you close the tab. DocBrain keeps every answer, links the entities into a connected memory, learns which sources to trust from feedback, and turns recurring questions into durable knowledge. The 100th person asking about a topic gets a faster, better answer because of the first 99. It's one shared memory for the whole organization with access control, not a private session per person.
How are docs quality-scored?
Three layers: Structural (deterministic), Style (configurable rules per team), and Semantic (LLM-assessed). Composite score 0-100.
Can I integrate with PagerDuty / incident management?
Yes. Native PagerDuty and OpsGenie ingestion. Knowledge Stream fires early warnings during incidents. Webhook subscriptions push events to any tool.

Security & Access

Can DocBrain change anything in my connected tools?
No. It's read-only, and that's enforced, not promised. When DocBrain connects to a live tool over MCP, any tool that could create, edit, or delete gets dropped during discovery. It can read Jira, Confluence, or Slack to answer a question. It cannot file a ticket, post a message, or change a page.
Will users see data they're not allowed to?
No. DocBrain enforces access at query time on two fronts. Source-system ACL mirrors the real permissions of Confluence, Slack, GitHub, and Jira. A page someone can't open in the source gets filtered out of their answers, side-channel-safe and audit-logged. For live tools, each user connects with their own OAuth, so live results are scoped to exactly what that person can see.
How is access controlled inside DocBrain?
SSO via GitHub, GitLab, or any OIDC provider. Role-based access control with four roles, space-level isolation between teams, and API keys hashed with Argon2. Every privileged action is audit-logged.
Can I run it fully air-gapped?
Yes. Self-host and point DocBrain at a local Ollama model. No API calls leave your network, zero data egress. Documents, queries, and embeddings stay entirely on your infrastructure.
How are secrets and credentials handled?
OAuth tokens are encrypted at rest and refreshed transparently; API keys are Argon2-hashed and never stored in plaintext. Connector credentials are referenced by name, never logged, and never surfaced in the UI. LLM and source keys live in your environment, not in DocBrain's database.
Is the source code auditable?
The source publishes in H1 2026 under BSL 1.1. Until then, you audit what actually determines your risk: DocBrain runs entirely inside your infrastructure from versioned Docker images and Helm charts you pin and control, with a published threat model covering 11 attack vectors, a zero-egress mode using local models, and enforced read-only access to your tools. Nothing leaves your network for you to trust us with — audit the boundary, not just the codebase.

Self-host in 5 minutes

Docker, an API key, and three commands. That's it.

quickstart
$ git clone https://github.com/docbrain-ai/docbrain
$ cd docbrain
$ bash scripts/setup.sh
✓ Created .env with defaults
✓ Started PostgreSQL, OpenSearch, Redis
✓ Ran 43 migrations
✓ Ingested sample docs
✓ Server running at http://localhost:3000
✓ Web UI at http://localhost:3001
Bootstrap admin key: db_sk_boot_..

License & community

Free to run. Honest about what's open.

DocBrain is licensed under BSL 1.1 and free for production use. The client tooling — the CLI and the IDE MCP server — is open source (MIT) in the repo, built and tested in public CI. The server source is not published yet — we missed our original H1 2026 target, and rather than quietly moving the date, we're telling you plainly: no new date until we can hit it. What's public today: client source, production Docker images, Helm charts, full configuration, and the threat model to run it yourself.