Private beta · summer 2026

Stop using AI.
Start running it.

Your AI workforce, organised like a real company — Chief of Staff to run your day, Project Coach to discipline your LLM, AI Staff to handle the loops. You're the CEO. Any LLM, every device, ~60% off flagship cost.

The Suite

Six capabilities. One Brain.

Each plug-in extends the kernel with a specific job — evaluate, attend, govern, retain, federate. Install one. Or all of them. They compose into something more than the sum of their parts.

What each does

Capability by capability.

Each plug-in is useful alone. Together they form a Brain that knows what you're working on, what's incoming, what's running, what you knew before, and what your team is doing.

core-brain
Foundation · LLM command center

The right LLM for every task. Same memory, every device.

Core Brain is your command center for working with AI. Connects you to multiple LLMs — OSS and commercial — through one interface, and keeps your memory portable across every device you work on. Use Claude for nuance, GPT for code, Llama for cheap bulk — switch any time, never lose the thread. Save time. Save money.

  • Multi-LLM by design — Claude · GPT · Gemini · OSS via Together / Fireworks / DeepInfra; pick per task or mix mid-conversation
  • Memory follows you — laptop, desktop, phone, cloud — same Brain, same context, encrypted in transit, owned by you
  • Save money on routing — default 50/35/15 OSS / cheap-commercial / flagship mix delivers ~60% off flagship cost at flagship-quality finish
  • CLI or desktop, or both — work in your terminal, in a native app, or alongside Claude Code / Cursor — same Brain underneath every surface
$brain chat # the command center, in your terminal
admin-brain
Attend · your Chief of Staff

Your Chief of Staff.

Admin Brain is your Chief of Staff. Email, calendar, tasks, projects, agent proposals — every stream that would normally pull you between fourteen apps — filtered, organised, prioritised, and responses pre-drafted, all in one place. Glance up, respond to ten emails, glance back down in two minutes. Your real work barely notices.

A great Chief of Staff doesn't just manage. She designs systems.

When she sees you doing the same work repeatedly, she proposes a Power Loop — a system or process for the work — and hands it to Agent Brain to turn into a human-in-the-loop workflow. Your morning support triage stops being a chore you do every day. It becomes a process that mostly runs itself, surfacing only what genuinely needs the CEO.

  • All your inboxes, one place — email, calendar, tasks, projects, agent proposals, team asks; a Chief of Staff doesn't juggle fourteen apps and neither should you
  • Filters & prioritises — surfaces only what truly needs the CEO; the rest gets handled in the background or batched for later
  • Drafts your responses — every reply already has a starting draft you can edit and send in seconds
  • Spots the patterns — proposes Power Loops when work repeats; hands them to Agent Brain to automate
  • Respects your focus — knows when to wait; the 2-minute clear-the-top pattern, not constant interruptions
$brain admin focus # 2 minutes. clear the top. back to work.
project-brain
Evaluate · your Project Coach

Your Project Coach. For you and your LLM.

Your LLM can write any document in the universe — and that's the problem. Without discipline, it'll happily build an entire app with no vision, no requirements, no design spec. Future-you ends up paying for that.

Project Brain is the opinionated coach that fixes that. It asks the one question your LLM never asks itself: "what's the evidence you did this right?" That you're solving a real problem. That you evaluated your options and chose wisely. That you understand your constraints. That you tested what needed testing.

Project Brain asks. LLMs answer.

You run it on demand today — an agent can schedule it tomorrow. It reads every doc you've built, gives your project a simple progress score, and suggests what you and your LLM should fix next. Every future LLM interaction gets sharper because there's a real spec to build on — and agents reference the structured state directly, without burning tokens grinding through markdown.

  • The evidence question — "what's the evidence you did this right?" pushed at every milestone; both you and the LLM have to show your work
  • Score & suggested fixes — run on demand today, scheduleable by an agent tomorrow; produces a simple score and concrete next steps
  • Opinionated, not blank-canvas — knows what a healthy project looks like (users, problems, goals, constraints, decisions, risks, tests) and coaches you toward it
  • Compounding context — every assessment makes future LLM work sharper; agents reference the structured state directly, no tokens burned on markdown
  • Travels with you — project state lives in Brain, follows you across devices and across LLMs; no project ever stranded in one tool
$brain project assess --all # the coach asks "where's the evidence?"
agent-brain
Govern · your AI Staff

Your AI Staff. Finally hire-able.

Most agent frameworks assume you already know what to automate and how to make it work. That's a strange assumption — if you knew, you'd already have done it. Agent Brain starts higher up the org chart.

Because the plug-ins compose, you don't just get agents — you get a real org chart: You (CEO) → Admin Brain (Chief of Staff) → Agents & Workflows (AI Staff). Project Brain knows what your projects are. Admin Brain spots the repeated work and proposes Power Loops — repeatable systems for getting it done. Agent Brain turns those into human-in-the-loop workflows: sometimes your step, sometimes a staff step, every iteration closer to the project being done.

You say the word. They do the work.

Example: an online product with a support inbox. Daily Power Loop = sort and prioritise the morning's emails, reply to the easy ones, kick off investigations on the hard ones, update docs where needed. Defined once. After that, it runs as a workflow — Agent Brain handles what you've approved for automation, surfaces the rest for a quick decision, and the support emails never pile up.

  • AI Staff, not just agents — agents and workflows you can task, supervise, retire, and rehire
  • Power Loops are the starting point — Admin Brain proposes the systems; you and Agent Brain build them out
  • Human-in-the-loop by design — sometimes you have the step, sometimes the staff does; either way you stay in command
  • Picks up from project plans — agents pull tasks directly from project-brain; no rebriefing every time
  • CEO controls always — approvals, spend ceilings, sandboxed execution, full audit trail
$brain agent loops # the Power Loops your Chief of Staff has suggested
Surfaces & interop

Your AI memory travels with you.

Use Brain as your AI workstation. Or stay in your favourite LLM tool and pipe Brain through it. Either way, what you learn, decide, and remember follows you — device to cloud to device, federated.

Workstation · terminal-native

brain-chat

A first-class AI workstation in your terminal. Type while the LLM streams; queue text for the next turn or hit ESC to interrupt. Pipes seamlessly into the rest of your dev flow.

  • Type-during-stream UX — the input stays live, queued or interrupting
  • Curated tool catalog — explicit opt-in for anything beyond safe defaults
  • Works alongside Claude Code, Codex, Cursor CLI, Goose, & co. (shared MCP)
Workstation · native desktop

brain-desktop

Three-pane native app for Mac, Windows, Linux. Chat on the left, Explorer in the middle, Viewer on the right. The same Brain as the CLI — every plug-in renders here, panes are aware of each other.

  • Pane-aware chat — the chat knows what's open in the viewer
  • Rich viewer — documents, briefings, agent graphs, project boards
  • Same web bundle serves brain.nerrem.com when you're on a borrowed machine
Interop · MCP standard

MCP Server

Brain speaks Anthropic's Model Context Protocol. Point Claude.ai, Claude Code, Cursor, Windsurf, ChatGPT Desktop (any MCP-capable client) at Brain, and your memory and tools are right there — bidirectionally. Queries pull Brain context; conversations save back.

  • Read-write — context pulled in, conversations captured back
  • Scope-respecting — every MCP call goes through agent-brain's policy edge
  • Tool surface — your installed plug-ins' tools become available to any MCP client
Interop · capture LLM conversations

Capture Gateway

Targeted capture from the LLM tools you already use — Claude, ChatGPT, Perplexity, and the other clients you live in. Your conversations across all of them flow into Brain so you can search across them later — and synthesise across them once data-brain ships.

  • Per-provider adapters — focused integrations for the AI clients you actually use, not a generic clipper
  • Searchable today — full-text across every conversation Brain has captured, regardless of which LLM it came from
  • Synthesis with data-brain — cross-conversation summaries and source-grounded answers arrive when data-brain ships

The federation thread: Brain lives where you do. Your laptop, your phone, Nerrem Cloud, a teammate's Team Brain — all the same Brain, encrypted in transit, signed gossip for replication. Whatever AI tool you're using, your memory's already there.

Better together

The plug-ins compose.

Each capability emits events. The others subscribe. The result is a Brain where decisions, attention, agents, memory, and team context all reference each other automatically.

A milestone you set in project-brain with a delivery date in 3 weeks shows up on admin-brain's briefing the morning of week 2 — "this is getting close, here's what's left."
A vendor invoice lands in admin-brain's inbox. agent-brain drafts the payment authorization using your standard policy — surfaces it for your single-click approval. The cost of a vendor email drops from 8 minutes to 8 seconds.
You hit 80% of today's token budget. core-brain quietly throttles agent-brain to OSS models for the rest of the day — the agents keep working, the quality stays acceptable, you don't get a surprise bill.
A teammate's team-brain announces a Friday commitment. Your admin-brain briefing the next morning surfaces "Alex is shipping Friday — anything they need from you?" — without anyone running a status meeting.
Principles

Four things we won't compromise.

Brain is architected so these stay true no matter how the suite grows. Every new plug-in, every new surface, every new integration — has to honor these.

01 · LOCAL-FIRST

Your machine is home.

The CLI runs locally. The desktop app runs locally. Even Nerrem Cloud is just a place to host your tenant — you can move it anywhere. Default to local; cloud is opt-in.

02 · OWNED

Your data. Your keys.

Passphrase-derived encryption. Per-tenant isolation. Export anytime in standard formats. If Nerrem disappeared tomorrow, your Brain would keep working.

03 · FEDERATED

Personal Brain first.

Team Brain is a federation of personal Brains, not a shared workspace you log into. Members own their data. The team layer is what's chosen to be shared.

04 · NOT YOUR DATA, NOT OUR BUSINESS

What you do with Brain is your business.

Brain is commercial proprietary software. Our business model is selling a product that works — not harvesting what you do with it. No training on your data, no usage fingerprinting, no quiet phone-home. LLM calls route through Zero-Data-Retention endpoints. (core-brain may go OSS down the road; the products around it stay ours.)

Coming next

On the horizon.

Two more plug-ins are designed and shaped — work to ship is ahead of us. They're slotted into the family so you know what's coming and how it fits.

Planned · v1 specified
team-brain
The team layer that doesn't own your brain

Small teams coordinate without giving up their Personal Brains. Each member stays sovereign — team-brain is additive, not a workspace lock-in. Shared goals, ratified decisions, and a team-owned knowledge layer live in a separate team scope the team agrees on, governed by three decision modes (unilateral, democratic, input+decider).

Decisions that survive the chat scroll. Comments-on-anything is durable, signed, and threaded — with permalinks that outlive team reorgs and member departures. The team's compounding knowledge artifact for years, not the next time someone clears Slack scrollback.

Federated by design — Nerrem can host the optional relay, or you don't use one. Your team could run on 100% OSS, on wholly-owned hardware, in an air-gapped facility, and still get the full team experience.

  • Personal Brain stays sovereign · team layer is additive (not workspace lock-in)
  • Ratified agreements that agents trust as team policy without re-asking
  • Comments-on-anything with permalinks that survive years
  • Cross-team unified inbox for members across multiple teams
  • 100% OSS option for air-gapped / regulated environments
Planned · being envisioned
data-brain
Precise LLM reasoning over your data

Databases are huge. LLMs are token-greedy. Today, asking an LLM to "analyse the data" usually means dumping a CSV into a context window and burning thousands of tokens before any real reasoning happens — or writing the SQL yourself and only inviting the LLM in once the result set is small.

data-brain inverts that. It separates the SQL from the LLM and gives you (and your LLM) a set of precise, composable commands for targeting the result set first, then reasoning on the small thing. The LLM can even drive the SQL it uses to look at the data — exploring, projecting, refining — before being asked for insight. Token minimisation over big datasets, by design.

Companion idea: once the LLM is reasoning over precisely-targeted slices, the natural next product is a living dashboard — visualisations that update as the LLM refines what it's looking at. Possibly its own product; definitely a direction we're shaping.

  • Separate SQL from LLM — target the result set first, reason after
  • LLM-driven exploration — it can drive the SQL it uses to view the data
  • Token minimisation by design — orders of magnitude cheaper on big datasets
  • Composable commands you can mix into agent workflows
  • Companion: living dashboard, driven by the same dynamic

Start with one. Grow into the suite.

Install core-brain first. Add the plug-ins you need, when you need them. Every one composes with the rest.