Agents don't need memory.
They need a Brain.

Knowledge that improves as fast as your agents work.

Connect one Brain to every agent you run, and they all work from the same current answer. Let them improve it on a branch: rewrite the pitch, restructure pricing, try a new escalation policy. Keep only what works, and every agent, teammate, and customer product has it the moment it lands.

Illustrative product walkthrough
NRevenue Operations Brain
AKSKJL+3
Connect
KnowledgeReview 3Activity
Go-to-market

ICP definition

Reviewed

Last reviewed May 14 · Alex Chen

We sell to operations leaders at 50 to 500 person companies in the Pacific Northwest. Northwind Pro is discontinued and must not be positioned in any new opportunity.

Reviewed changeMay 14 · Alex Chen

Target market narrowed from national to Pacific Northwest. 47 prospects archived in the same change.

Source lineageOpen sources
  1. 1 Q3 board deck
  2. 2 CRM territory export
  3. 3 Revenue approval
Team workspace

Build, review, and use knowledge together.

Desktop agents

Connect Claude Code, Codex, OpenCode, or any MCP client.

Embedded products

Connect through API or MCP without creating another silo.

Start now.

Do it yourself in a browser, or hand it to the coding agent you already run. Both end at the same Brain.

Do it yourself

Create a Brain and drop a folder or a .zip on it. No terminal and no agent. Import is a copy, so your existing tools keep working exactly as they do today.

Start now What import does

Have your agent do it

Copy one and paste it into Claude Code, Codex, OpenCode, or any coding agent. The prompt is the whole procedure, so there is nothing to install first.

Step by step for each one

Two agents and four tools is fine. Then it isn't.

Notion, Docs, Drive and Confluence are editing surfaces built for people, and they are good at that. Their unit is a page, their API is designed around a person clicking, and nothing in them carries a typed field, a review step, or a record of why something changed. An agent pointed at them is reading a layout to guess at facts.

They never held the truth. They held documents about it, and a person did the reconciling. Hand that job to agents and the reconciling does not disappear, it just happens faster, in more places, with nobody watching. A document is a fine rendering and a bad database.

Wiring N agents to M tools is N × M links to build, authorise and keep correct. Routing them through one Brain is N + M.

Week 1One person wires up one agent

Three tools, three integrations, and it works. This is the demo everyone gives.

31 × 3 connections
Agents on the left, tools on the right. Every line is something a person has to build, give a credential to, and keep correct.
  • Read from anywhere. Accept in one place.Knowledge is still born in Slack, email and meetings, and your agents can bring all of it in. Capture is many to one. Authority is one place. That is the part a mesh cannot do, because a mesh has no center.
  • Trying costs you nothing.An import reads; it never writes back. No OAuth scopes into your systems, no live coupling, nothing to unwind. Run both for a couple of weeks and compare the answers.
  • Then cut over, on purpose.A copy drifts the moment people keep authoring somewhere else, so this is a trial with a shelf life, not a resting state. When the Brain is the better answer, make it the place you write.

Agents that change things, not just answer questions.

Ask in plain language. The agent resolves it against typed records, not a pile of documents, and stages the whole batch as one change. Approve it once, and the moment it lands every surface that reads this Brain is current.

See how a change lands
Agent task
chg_02b41c
Archive every prospect outside the Pacific Northwest. We are changing our focus.
47 prospects matchedQueried on region across 1,284 records
Staged for review
RejectApprove all 47

Change it once. Every agent knows.

Memory is private and passive: each agent keeps its own copy, and nothing it learns reaches anyone else. A Brain is one body of knowledge your people and agents build together, structured so an agent acts on records instead of guessing from prose. Improve it in one place and everything that reads it is current.

BeforeScattered knowledge. Different answers.
Sales agentYes, pitch Northwind Pro.
Support agentOnly for enterprise.
Marketing agentIt is discontinued.

Three sources. Three answers. One of them reaches a customer.

AfterOne reviewed Brain. Cited answers everywhere.
N
Reviewed and currentProduct catalog · accepted May 14
Sales agentDiscontinued. Offer Atlas.Product catalog
Support agentDiscontinued. Offer Atlas.Product catalog
Marketing agentDiscontinued. Offer Atlas.Product catalog

One change. Every agent current the same minute.

Try it on a branch. Keep what works.

Open a branch and let agents rework the knowledge in parallel: a new pitch, a different pricing structure, a rewritten escalation policy. The version everyone else reads does not move. Propose the changes that earned it back to the main Brain and leave the rest on the branch.

Code got fast when changing it stopped being risky. Knowledge gets the same loop.

See how a proposed change lands
Branch · pricing-v2
3 agents · 14 changes
Pricing exceptionsBranched from the main Brain · May 20
  • Annual prepay discount capped at 15% Propose
  • Nonprofit pricing folded into the Team tier Propose
  • Usage-based overage for API seatsLeave on branch
Keep workingPropose 2 changes

One Brain. Every way your organization works.

The same Brain your team edits in a browser is the one your desktop agents connect to and the one your product provisions per tenant. Nobody has to choose between a polished product and an extensible platform, and nothing has to be migrated when you move between them.

Workspace

Work together in Nicia

Create pages, organize sources, invite teammates, review proposed changes, and inspect activity without dropping into code.

Open the workspace
Connection

Connect desktop agents

Give the agents people already use access to reviewed knowledge, with scoped Connections and visible provenance.

Connect an agent
Deployment

Build it into your products

Provision Brains for your tenants, query typed state, and return cited context through HTTP or MCP. Point your coding agent at nicia.ai/build.md and it wires the whole path.

Read the activation guide

Let agents do the work. Keep the last word.

Review is not a brake on your agents, it is the reason you can turn them loose. Every change arrives with its intent and its evidence. Approve the exact assertion or reject it, and what you accept stays connected to the source it came from.

See how Review works
Proposed change
chg_019f7a
Product catalogProposed by Sales agent · 9m ago
Needs review

Northwind Pro — available, all segments.

Northwind Pro — discontinued. Offer Atlas.

RejectApprove change

Every answer can show its work.

A Brain returns context with citations and a receipt for exactly what was consumed. When an answer turns out to be wrong, correct it once and list every read that was served the old value. Teams inspect the same provenance in Nicia; products carry it into their own experience.

Explore receipts
Revenue Operations Brain Context ready
Who are we selling to right now?
Operations leaders at 50 to 500 person companies in the Pacific Northwest. Northwind Pro is discontinued.
[1]
ICP definitionReviewed May 14
Receiptrcp_019ffefe
Knowledge observability

Find out what your agents are actually reading.

Every answer an agent reads through a Connection is recorded with the exact sources behind it, pinned to the versions the reader saw. Open any one of them and run a context check: it reports where the sources contradict each other, where the same passage was sent twice, and where you paid for words that carried nothing.

Import a copy of what you have, point one agent at it, and you can check the first answer it gets.

See how reads are recorded
support-agent7 sources · today
Risk · contradiction

One source says Northwind Pro is available in all segments. Another says it is discontinued.

icp-definition · product-status
Cost · repetition

These two sources share 84% of their wording. The agent was shown both.

q3-pricing · pricing-review
Checked 7 of 7 sources.
For the people shipping the AI product

You already built a worse version of this.

Every team that gives its customers their own context ends up writing the same four things badly. We know because they keep describing them to us in the same words.

“Our RAG returns chunks, not answers.”
Ask what the current ICP is and a vector store hands back five passages from three different quarters, all of which retrieve equally well. Nothing in it knows that March was superseded in August. A Brain keeps typed records with corrections, so current is a property of the data instead of a prompt-engineering problem.
“Onboarding ingest is our worst code.”
Customers upload their materials at signup and you pull out a company description, an ICP, a product list. You wrote that once, you hand-tune it constantly, and it surprises you on every new customer. Send the documents to a Brain and get back cited statements, each traced to the passage it came from, ready to show the customer for confirmation.
“We are scared to change the schema.”
You did not plan for the model to evolve, so every new field is a migration and a backfill and a bad afternoon. Declaring a field in a Brain is retroactive: nothing you ever sent was dropped, so promoting a key makes values you already stored declared, returned, and filterable without touching a record.
“Every customer needs their own.”
A git repo cloned per customer, or a tenant column and some hope. Provision a Brain per tenant addressed by your id, with its own keys, its own review queue and its own export. No Nicia ids to store and nothing to reconcile.

The product your team uses. The platform you build on.

Start in Nicia, connect the desktop agents your team already uses, or provision one Brain per tenant from your backend. All three paths meet at the same review, policy, and provenance boundary.

Open the activation guide
read-context.sh
curl -X POST "$API/v1/brains/$HANDLE/context" \
  -H "authorization: Bearer $AGENT_KEY" \
  -H "content-type: application/json" \
  -d '{
    "prompt": "Who are we selling to right now?",
    "consumerId": "sales-agent",
    "format": "text"
  }'

# cited context + consumption receipt

One per customer. One per project. One per employee.

A Brain is the unit, and it is the one you already bill on. Review, receipts, the API, MCP and full export are in every tier including the free one, because a Brain nobody can audit or leave is not worth having.

Free
$0 / month

10 Brains included.

  • API and MCP
  • Review queue and receipts
  • Full export, any time

Enough for a team, a handful of clients, or a product in development.

Team
$99 / month

Plus $40 per Brain, up to 50 Brains.

  • Everything in Free
  • SSO and roles
  • Approval policies per Connection
  • Audit log export
  • Managed ingest

Agencies and consultancies with a Brain per client. Companies giving every department its own.

Scale
$499 / month

Plus $12 per Brain, up to 500 Brains.

  • Everything in Team
  • Fleet provisioning
  • Priority support

Products that give every customer their own Brain. Companies that give every employee one.

Fleet
Talk to us

Thousands of Brains and up.

  • Everything in Scale
  • Isolated storage per Brain
  • Dedicated throughput and raised rate limits
  • Hosted white-labelled Brain UIs for your customers
  • Templates and bulk provisioning
  • Change feeds into your own pipelines
  • Per-Brain usage rollups you can rebill
  • Migration support and an SLA

AI companies reselling a Brain per end customer. Anyone whose fleet outgrew our limits. Talk with us

Leaving is a feature

You arrived with a folder. You leave with a model.

Export any Brain and you get plain markdown with your frontmatter intact, a catalog, a greppable activity log, revision history, and an AGENTS.md that teaches any agent how to read the whole thing. No proprietary envelope, on every tier including the free one.

See what an export contains
Talk to us

See it against your own knowledge.

Bring a folder, a wiki export, or just a description of what your agents keep getting wrong. We build the Brain on the call, point an agent at it, and you watch it answer from your own material. Thirty minutes, no deck.

You keep the export either way, whether or not you buy anything.

Your agents are ready for real work. Give them somewhere to do it.

Start with your team, your agents, or your product. Every path builds the same thing: one Brain every agent reads from, that your agents improve and your people steer.