New in June: The Librarian is live

AI Employees

Knowledge-Ops

In development Does the work of a research librarian

Every answer with receipts: one grounding layer over your wiki, vectors, graphs, and files, so your AI cites company truth instead of inventing it.

Follow the build log Pair it with Center Deep Research
Your knowledge stores: wiki, Qdrant vectors, FalkorDB graphs, files, metricsKnowledge-Ops grounding layer: The Librarian + kb.ground with local embeddings and inferenceCited answers out to people and agents: web UI, REST, and MCP

Who it is for

Company knowledge that answers with citations.

Teams running AI agents who are done with confident wrong answers: agent builders, ops-heavy companies, and anyone whose institutional knowledge is scattered across wikis, drives, and databases.

Why it wins

Grounded answers. Not confident guesses.

01

Your agents stop hallucinating company facts. Every answer names its sources: the wiki page, the graph node, the actual file.

Answers company questions with cited evidence, blending your wiki, Qdrant vector collections, FalkorDB knowledge graphs, indexed files, and metrics in a single query

02

No migration project. It federates the knowledge stores you already have instead of forcing you to copy everything into yet another silo.

Runs a living internal wiki with an edit UI, full doc history, and API upsert, so SOPs and runbooks become machine-groundable the moment they're written

03

Everything runs local: embeddings, vector and graph stores, and inference on-prem, never a per-token cloud bill: local embeddings, local vector and graph stores, local inference, and single sign-on at the front door.

Serves your whole agent fleet over MCP with 10 JWT-validated tools, so any agent can ground against company truth before it opens its mouth

How it works

Federate stores. Ask The Librarian. Cite.

01

Connect what you already have

Point Knowledge-Ops at what you already have: your wiki, vector collections, knowledge graphs, file shares, and metrics. It indexes and federates, it doesn't duplicate.

02

Ask, or let your agents ask

Ask in the web UI, or let your agents ask over REST and MCP. The Librarian, our dedicated grounding agent, fields questions the same way a great reference librarian would.

03

Get answers with receipts

Get an answer with an evidence trail: which sources were searched, what was found, and exactly where each claim came from. Weak evidence gets flagged, not papered over.

Screenshots

Product tour coming soon.

Ask view showing a grounded answer with source citations

Product capture coming: Ask view showing a grounded answer with source citations.

Wiki editor with a runbook open and full doc history

Product capture coming: Wiki editor with a runbook open and full doc history.

Live source-status cards and metrics charts on the dashboard

Product capture coming: Live source-status cards and metrics charts on the dashboard.

Proof and access

What's proven today.

6+source types federated in one query
10MCP tools, every call JWT-validated
1,179nodes in one live knowledge graph it cites

Proof points

  • Live at knowledge-ops.magicunicorn.dev behind Unicorn Commander SSO. We run our own company's knowledge on it every day.
  • Federates 6+ source types in one query: wiki, file index, Qdrant vector collections, FalkorDB knowledge graphs, metrics, and Center Deep research.
  • Verified grounded answers cite multiple live sources at once, including a 1,179-node knowledge graph and a 396-point vector collection alongside wiki docs.
  • Runs entirely on our own hardware: local bge-m3 embeddings and local Qwen inference on a consumer RTX 3090. No per-token cloud bill.
  • Our Brigade agents (The Librarian, our lights-out datacenter agent, our marketing orchestrator) ground against it daily.

Pricing and access

Live and in daily production use across our own agent fleet. External signup isn't open yet; follow the build log and you'll see it open.

Shipping log

What just shipped.

  • June 2026: shipped to production with a web UI, HTTPS, and Unicorn Commander SSO
  • June 2026: The Librarian went live as a Brigade agent with its own generated avatar and voice
  • June 2026: infrastructure runbook corpus seeded; our lights-out datacenter agent now grounds hardware fixes against it

FAQ

Plain answers.

What does it actually search?

Everything you point it at, in one pass: your internal wiki, indexed files, Qdrant vector collections, FalkorDB knowledge graphs, recorded metrics, and Center Deep research. One question fans out across all of them and comes back as a single cited answer.

Do we have to migrate our data into it?

No. Knowledge-Ops federates the stores you already run instead of duplicating them. Your wiki stays your wiki, your vector DB stays your vector DB. We just make them answer as one.

Who is The Librarian?

A dedicated grounding agent with its own face and voice. Ask it a company question and it answers with cited evidence instead of a confident guess. When the evidence is thin, it says so.

Can our own agents use it?

Yes, that's the point. It exposes 10 MCP tools registered with our Brigade agent broker, every call JWT-validated, so any agent in your fleet can ground its answers before acting. There's a REST API and a web UI for humans too.

Is it available today?

It's live and running our own company's knowledge every day, and our agents ground against it around the clock. External signup isn't open yet; we'll open it up as capacity allows.