Closed beta. Invites go out in waves.

One person.
Every role.

A working team of AI agents on a server you own. One memory that compounds. Run it from your phone.

See it run: seven views of a box in production

This form needs JavaScript, because the spam check it posts through does. With JavaScript off, email support@agentchakra.com and we will add you by hand.

Seven components. Watch each one work.

Views of the founder's own box, captured from the running console.

The Agents view: six agent cards named by role, Personal, Founder, Business Development, Engineering, Content and Chief of Staff, each with live instruction, skill, tool and channel counts. — Agents Meet the staff: six agents, each with a role, its own skills, and live work counts.

The Skills catalogue on mobile: 241 skills, 132 active, grouped by domain. — Skills A search recorded live: 241 skills narrowed to 11 as the word is typed. The Skills catalogue: 241 skills, 132 active, grouped by domain.

The Connections view: 29 connected tools with per-tool descriptions and agent access counts, above a catalogue of 1,045 apps. — Connections Your apps, reachable: 29 tools connected here, from a catalogue of 1,045.

The Channels view on mobile: connected channel, catalog, and rooms with policies. — Channels You talk to them where you already are: six rooms, live policies, member counts.

The Brain dashboard on mobile: stacked stat cards with live knowledge counts. — Brain Everything they learn is kept: 19,967 pages, 10,049 facts, and a living knowledge graph. Everything they learn is kept: 19,967 pages and 10,049 facts, counted live.

The router console on mobile: the provider catalogue with live connection counts. — Router One router, every model: 21 live providers from a catalogue of 256, grouped into failover combos. One router, every model: 21 live providers from a catalogue of 256.

The Automation view: a registry of 86 scheduled jobs, 77 active, each showing cadence and last run time. — Automation 86 scheduled jobs, 77 live, each on its own cadence. The boring parts run themselves.

The problem

The bottleneck was never the AI.

The models are extraordinary. The way most people have to use them is not.

AI forgets you every morning.

Chat AIs meet you again every morning. You re-explain your business, your clients, your taste, and the answer is good for exactly one tab. Close it, and the understanding is gone.

Your tools scatter your context.

So you add tools. Each one brings a login, a subscription, a policy that can change under you. Your context scatters across a dozen apps, and you become the only thing connecting them.

Your work lives on someone else's terms.

Everything you build lives on someone else's servers, under someone else's rules. If a provider raises prices or shuts a product down, your way of working goes with it.

The real limit is your attention: one brain, one thread of it, every role competing for that thread. You need a team that carries your judgment, and a place of your own to run it.

Built for a specific kind of person.

Agent Chakra is for people who already run something and refuse to accept that growing it must mean drowning in it.

Founders and operators who want a team's output while staying one person to the world.
Business owners who want to multiply the people they have, not replace them, and to keep only the work that sits in their zone of genius.
Builders without a company yet who want the distance from idea to running business crushed.
People who work from anywhere who need their business to run as well from a phone on a beach as from a desk.

It is not for everyone.

If you love operating the machinery yourself, the open-source agent frameworks are excellent, and we mean that without irony. Build there. If you later want it run for you, bring your agent in; it keeps what matters.
If you want an AI that needs nothing from you, this is not that. Your judgment, taste, and approvals are the point. The system carries the work; the direction stays yours.
If a chat window already covers your needs, keep it. This is for when the chat window stopped being enough.

Three things, recorded as they happened.

One is read live out of this site's own repository when the page is built; the other two are screen recordings of Telegram threads with the running agents. Each film discloses on screen every cut, speed change, and cover, and the threads carry their [DEMO] label as recorded.

This page was built by the team it describes.

It was written and shipped by our own agents, with the founder approving every round. They work on top of the frontier models, exactly as yours would, which is why a model's name sits in the co-author line under this log. The log is the repository's own, read live when this page was built, that line and all.

$ git log --oneline -8
1072839 fix(k14 round7 pass B): fold the served GPT-5.6 audit findings
33cdfa4 fix(k14 round7 pass A): fold the served Fable audit findings
9bb804d k14-website-wow-round-20260730: film presentation, hero focus, a11y pass (round 7)
2ed3b34 k14-website-wow-round-20260730: copy and product-truth pass (fix round 2)
45278dd Keep the checkout copy to what the film shows
2980e2a Say who built this page, and what it was built on
d8ecbe6 Load the smaller video file first in the product reel
4277067 Make the films' screen-reader descriptions match their captions

$ git log -1 --format=%B 1072839 | grep Co-Authored
Co-Authored-By: Claude Opus 5 <[email protected]>

Read at build time, so it is current with the page.

One ask, a whole team.

Give the chief of staff one request with two jobs inside it and it splits them: this piece to the business agent, that piece to engineering, each on its own run. The film in this card is one of those runs as it happened. Two delegations go out. One comes back with a number. The other is still working, so the answer says so instead of inventing one.

This recording will not play in your browser. Download the file, or open the shot-by-shot transcript below.

A run we asked for, recorded in Telegram on 31 July 2026 off the running system. Two delegations go out with their own run ids; the number that comes back is read out of the connections catalog and carries the catalog's own as-of time, and the second run is reported as still pending rather than answered. The single speed change is labelled on screen.

Read the film instead, shot by shot

The request goes in at 11:40 and is put again at 11:46.
Both delegations are accepted, each with its own run id: b17b6272 to the business agent, fdf1552a to engineering.
The chief of staff declines to invent the two numbers while they are still out, and the routing card underneath shows which model took the work and how long it thought.
A status check is typed and sent at 12:41.
Seven minutes of real waiting, played at 110x and labelled on screen.
The answer lands: nine connected apps in that demo workspace back from the business agent, engineering reported as still pending rather than answered.

The purchase that needs you.

An agent can walk a checkout all the way to the payment screen. It cannot cross it. Every purchase waits for an approval that only your phone can give. Full autonomy up to the money. A hard human gate at the money. Below, that gate holding: a purchase request, worked to the end, stopped at the approval, answered with a no, and stood down with nothing bought.

This recording will not play in your browser. Download the file, or open the shot-by-shot transcript below.

A run we asked for, recorded in Telegram on 31 July 2026 end to end. Nothing was bought. One payment line in the agent's report is covered by a labelled bar we added in edit, and the film says so on screen while it is covering it.

Read the film instead, shot by shot

The purchase request is typed and sent on camera at 12:53.
Ten real minutes of the agent working the request, played at 46x and labelled.
The agent's own report lands and stops at the approval gate. Its payment-method line sits under a labelled bar we added in edit; the label above the bar stays readable, so you can see what was covered.
The denial is typed and sent on camera at 13:08.
A second stretch of waiting, played at 10x and labelled.
The stand-down arrives with its routing card. Nothing was bought.

What Agent Chakra is. And what it is not.

Agent Chakra is

An agent operating system: a dedicated server, yours alone, with a working team of AI agents already on it. We run it for you.
A method. Seven components, arranged the way months of daily operation proved they should be. We call it the way of Chakra.
Yours to keep. Free trial in; a full way out, spelled out in Sovereignty below.

Agent Chakra is not

Another chatbot. Your agents hold roles, remember everything, and keep working when the chat is closed.
A rival to the AI companies or the agent frameworks. We ship an open agent engine and run it for you; the models stay whichever ones serve you best.
A platform that owns your work. Your data, your memory, your server.
A kit you assemble. It arrives working. Setup takes minutes, guided by your first agent.

Picture the open-source agent frameworks as a factory. Every machine you could ever need is on the floor, and the doors are wide open. The catch: you walk in alone, and working out what connects to what is your problem.

Agent Chakra is walking into that same factory after a pathfinder spent months inside with your exact goal. You inherit the findings. The configurations that work. The experiments that failed, so yours don't have to. The trail is already marked.

Your subscription is access to that trail, plus every finding the members after you add to it. And if you ever want your own keys to the factory, that door exists too; Sovereignty below spells it out.

Walking in alone

Walking in after the pathfinder

The method

Seven components. One system.

In the chakra tradition, well-being is seven centers working as one body. Same here: seven components, each one replaceable, none of them the system on its own. We call it the way of Chakra. Here is what happens when one message arrives.

Agents

The right agent answers.

The team starts as six seats: personal, founder, chief of staff, business development, engineering, and content. It shapes to the roles your business actually has. Each agent keeps its own voice, its own scope, its own tools, and to the outside world there is one identity: yours.

Skills

It knows how you do things.

Your procedures, written down once and improved with use. How you quote, how you write, how you follow up. Skills are portable knowledge that moves with you.

Connections

It reaches the tools you already use.

Over 1,000 apps in the catalogue: email, calendars, docs, CRMs, the rest of your stack. Custom connections can be brought in too.

Channels

It lives where you already talk.

Telegram first; Slack and Discord for teams and communities. No new app to learn, no dashboard to keep open. Work arrives in chat; approve, edit, or reject in one tap.

Brain

Nothing you tell it is wasted.

One organized memory per workspace that grows as a side effect of working. Decisions, people, projects, corrections: captured with sources and dates, searchable by every agent in that workspace, compiled into pages you can read. Your personal workspace stays walled off from your business ones. This is the asset. It compounds.

Router

The right model for each job.

Many AI providers sit behind one routing layer. Simple work goes to cheap or free models; hard work gets the best available brain. When one provider has a bad day, the router routes around it.

Automation

It works while you sleep.

Scheduled and background processes run around the clock, with a heartbeat that watches for what matters and only wakes the expensive thinking when it should. You review finished work in the morning instead of doing it at night.

Every message you send travels this path, engaging the pieces its task needs. Every run makes the next one better.

The parts nobody demos.

Autonomy is easy to demo. This is the machinery that makes it safe and keeps it running.

A vault your agents cannot read

Your accounts live in a locked vault. When an agent needs one, the vault swaps in a single-use pass, so the agent works without ever holding the raw password.

Watch their hands

Any agent working in a browser can be watched live from your screen, and one click pauses it and puts your hands on the same page.

Browsers, logged in as you

Agents work the web the way you do: logged in, on the sites you use, filling the same forms.

Money moves only through you

Every purchase stops at your phone for approval; an approved one mints a card that works exactly once. The film above shows that gate holding.

Meetings become memory

Record a call from your phone or desktop and the transcript lands in the Brain: speakers, dates, and decisions, so the team is briefed without you briefing them.

Your keys, provider rates

The models run on your own provider keys at provider rates. You see what intelligence costs, with no markup in between.

It maintains itself

Scheduled jobs watch health, refresh model catalogs, verify credentials, and self-heal drift around the clock. Updates arrive tested, with rollback ready.

Answers carry receipts

What the Brain knows is stored with sources and dates, and agents answer with citations you can follow.

Changes wait for approval

The system can propose its own improvements from feedback, but structural change lands only through your approval, onto an append-only record.

From invite to working team.

Your server wakes up.

We provision a dedicated server for you. Your name is on it, and your agents are the only ones running on it.
Your first agent says hello.

Setup is a conversation. Your first agent walks you through connecting your channels and tools. It takes minutes.
The team takes shape.

Agents come online around the roles you have. From then on, every day of use makes the system more yours.

The invite is the only step we cannot do for you.

Join the waitlist

Sovereignty

Yours means yours.

The people who get the most out of AI this decade will be the ones who own their setup. Not rent it. Own it. An AI woven into your business is too important to hang from any one company's pricing page.

Your capital is not any tool. It is your data, your judgment, your accumulated context: what you know, how you decide, what you would never do. Agent Chakra makes that capital concrete, organized, and portable. Most people believe they already have this. Almost nobody does.

So every component is replaceable by design. Models can be swapped. The engine is open. The channels are adapters, not foundations. If the market shifts, you re-point one part and keep working. The system is built to outlive any vendor's decisions, including, frankly, ours.

The way out is part of the product

You come in through a free trial, and membership runs month to month for as long as you want to keep testing the water; nothing asks for a long commitment. You can leave with everything: a perpetual license, self-hosting on your own hardware, and a full export of your workspace, your files, and your agents' entire memory. The full export lives in your account view: one click assembles it on your server and hands you a plain archive that opens anywhere. A system built on the rejection of lock-in does not get to lock you in.

What we won't pretend

When you use a frontier model, that provider's usage policies still apply. We think that is a fair trade for peak capability, and the router means no single provider is a point of failure.
A dedicated server costs more than a light user's pay-per-use chat subscription. Sovereignty is not the cheapest option. It is the one that is still standing when circumstances change.
A subscription with us is also a dependency. That is exactly why the way out above is part of the product.

Security, spelled out.

Sovereignty is a promise about power. This is the plumbing under it, stated plainly.

Where your system lives

A dedicated server provisioned for you alone, hosted in the EU by default. Your agents, files, and memory live on it; nothing sits in a shared database next to other customers.

How secrets are held

Credentials are sealed in encrypted envelopes before they touch storage and are write-only from the interface: you can rotate or delete a key, never read it back out. They are never printed to a log or a command line, and agents receive single-use passes, not the raw secret.

What gets backed up

Your agents' memory is dumped nightly on your own box, and the credential store makes a daily encrypted backup that only your recovery kit can open. Not us, not the hosting provider, not even the box itself can read it.

Who is watching the agents

You are, whenever you want: any agent browser session can be watched live and taken over with one click, and no purchase completes without an approval on your phone.

What we learn about visitors

Sites in the system run consent-first analytics: named product events only, no session replay, no cross-site identity, and first-party visitor events are kept for two months, then dropped. Analytics-provider retention is detailed on the privacy page.

How you leave

One export assembles everything on your server: your agents' memory as readable files plus a database copy, your workspace, skills, and the sealed vault backup. Ask for deletion and we confirm when it is done.

The credentials screen of the running product with the one-time secure link panel open. The stored keys listed behind it show a name and a status only, never a value. The panel explains that the link can be used once and expires in about fifteen minutes. — Handing over a key without ever seeing it. Whoever holds the secret pastes it on a private page and it lands straight in your vault. The link works once and expires in about fifteen minutes. Captured from the running product.

The fine print lives on the privacy page, written in the same plain language.

Receipts from the running system.

Andrea Ercolessi Founder of Agent Chakra I built this for myself first. The whole argument is in the manifesto.

The Agent Chakra console, Agents view: six agent cards named by role, Personal, Founder, Business Development, Engineering, Content and Chief of Staff, each showing live instruction, skill, tool and channel counts. — The console behind that receipt: my own six agents, captured live from the running product. A screenshot, not a mockup.

That console is what your invite turns on.

Join the waitlist

Questions, answered straight.

Why not just use ChatGPT, Claude Code, OpenClaw, or Hermes?

Keep them. Agent Chakra does not replace any of these, and it does not compete with them.

ChatGPT and Claude are chat interfaces: brilliant in conversation, but the understanding lives in a tab, and nothing happens when the tab is closed. Claude Code and Codex are serious engineering tools that assume you are comfortable in a terminal. OpenClaw and Hermes are open agent frameworks: all the machinery is there, and assembling and operating it is up to you.

Agent Chakra is the agent operating system on top of that machinery. We ship an open agent engine and run it for you, on a server of your own, arranged into a working team by the Chakra method, reachable in the chat apps you already use. If you have already built an agent in Claude Code, Hermes, or elsewhere, you bring it in and it keeps what matters. Nothing you use today has to lose for this to work.

Do I need to know how to code?

No. Setup is a guided conversation with your first agent, and daily use is chatting in Telegram, Slack, or Discord. If you do have technical skills, there is plenty of depth to go find. None of it is required.

What does "a dedicated server" actually mean?

A real server, provisioned for you alone. Not a shared platform where your work sits in someone else's database next to everyone else's. Your agents, your memory, and your files live on it, we keep it running and updated, and if you ever choose the self-hosted path, the same system runs on hardware you control.

What do I need to bring?

Your own model access and your own accounts. You connect the AI providers you already pay for, and the router spreads work across them at their rates, with nothing marked up by us. You connect the tools you want your agents to reach, and you pick the chat app you already use: Telegram, Slack, or Discord. The server, the agents, and the method come from us.

What happens to my data if I leave?

It leaves with you. A full export of your workspace lives in your account view: your agents' memory, your files, everything the system knows, assembled on your server into one archive that opens with standard tools. Beyond export, there is a perpetual license and a self-hosting path, so leaving our subscription never means abandoning your system.

What if an AI provider raises prices, changes rules, or disappears?

You keep working. Models reach your agents through a routing layer that spans many providers, so a price change, policy shift, or outage at one company means the router leans on the others. That resilience is a founding design law of the system, not a feature we added later.

How much does it cost?

Pricing is shared with your invite, before you commit to anything. The shape of the deal is settled: every membership starts with a free trial, runs month to month, and carries the exit terms described in Sovereignty. The number is not settled yet, so we publish none during the closed beta rather than post a figure that moves while the cohort is still being sized.

When will I get access?

Invites go out in waves, in waitlist order. The beta cohort is deliberately small, and each wave is sized by what we can onboard well, not by a growth target. Your confirmation email is your place in line; we will tell you when your wave is next.

How is this different from hiring a consultant to set up AI automations?

A consultant engagement leaves you dependent on the person who built it. Agent Chakra is a product: the setup knowledge is built in, the system maintains itself, updates arrive tested, and improvements discovered across the member base flow to everyone. You are buying a marked trail.