The Nagent platform
The platform under the growth team. One memory, one record, one ladder of trust. The infrastructure that lets agent teams do commercial work beside your people, and lets you see every inch of it.
- ISO/IEC 27001:2022 certified
- Every frontier model, routed per agent
- Eight hundred connectors, granted one key at a time
- Every action on an append-only record
What is the Nagent platform?
The Nagent platform is the infrastructure under the growth team: one memory, one record and one ladder of trust for AI coworkers working beside your people, which Nagent calls Multiplayer AI, with agent teams for marketing, sales, content and customer experience on top. Every action waits for approval until the agent earns trust, and it deploys in your private cloud.
Last reviewed
Built in the right order
Three layers, bound by one room. Agentic work needs the infrastructure that governs agents, the agent teams that do the commercial work, and the operating crew that makes the outcome land. Nagent builds the governing layer first, because the other two stand on it.
Multiplayer AI is what binds them.
People and agents work in the same room, on the same number, with every action on one record. The infrastructure provides the room and the record; the application layer staffs it; the operating crew sits in it until the record says the agents can. Remove the room and the three layers are three products. Keep it and they are one team.
- Agentic infrastructureThe platform. Organisation context, the shared workspace, orchestration, composable agents, skills, tools, models, memory, self-improvement, and the control plane over all of it. Four patents filed on the orchestration and memory methods, three published.
- Application layerTeams of named agents with remits and reporting lines: marketing, sales, content and customer experience, each with a chief who owns a number and specialists who execute against it.
- Operation layerForward deployed engineers who onboard the organisation and stand the system up, then trained operators who sit in the room with the agents and run it to an agreed outcome.
02 of 06
People and agents in one room
Every click becomes an agent action. A team room holds the people and the agents who share a number. A request is an instruction to an agent; the agent proposes the action on a confirm card in the thread; the person who owns the number approves it; the result lands where the next colleague can pick it up.
- Team roomsOne workspace per team, with the chief, the specialists and the people in it. Join a task in progress, hand it off, leave an instruction the agent keeps.
- Confirm cardsApproval where the work is, not in a separate queue. The click, the approval and the outcome are on the record together.
- Meetings and briefsBrief a team in its room, hold a meeting with it, and read the transcript and the actions beside the work.
03 of 06
Governance you can see into
A control plane over every agent, not a dashboard beside them. For every agent: its triggers, model, memory, skills, trust, autonomy, tools, budget and surfaces, each with an owner and a history. Every dimension is inspectable and enforceable, per agent, not per fleet.
- Roles for both kinds of colleagueAdmin, operator and viewer, scoped across people and agents alike. What an agent may see is set at organisation and team level and revoked in one click.
- Budgets and a kill switchDaily and monthly caps in currency, enforced by the runtime; a breach stops the agent, and so does a person.
- Tenant isolation and an append-only auditClient data stays in the client’s tenant. Every agent decision and action is written to a record the client can read and export.
04 of 06
Bounded autonomy
Autonomy is a level, and each level is earned. Every agent starts with every action waiting for a person. A trust score out of 100, rebuilt nightly from five signals, recommends the next level; a person signs it. Deleting and governance changes stay with people at every level.
- L1 ObserveReads and reports. Anything it would write, even a draft, waits for a person.
- L2 AssistReads and drafts on its own. Changes to records other people rely on wait for a person.
- L3 OperateChanges workspace records on its own. Anything that reaches outside the workspace waits for a person.
- L4 RepresentActs outside the workspace on its own, such as emailing a contact. Publishing waits for a person.
- L5 LeadPublishes on its own and can hand work to agents below it. Deleting and governance changes always wait for a person.
05 of 06
A learning loop
Measured against the number, not the deliverables. An agent is scored against the number it was hired for. Feedback from the room changes how it works without retraining; a change is evaluated before it reaches production; the mix of people and agents shifts as the record grows.
- Feedback that sticksAn edit, an approval or a correction in the room becomes a rule the agent keeps.
- Skills with a weightAttached procedures, each with a priority, so repeat work gets the approach that worked last time.
- Evidence before promotionWork approved without edits, risky calls handed up and no breaches over a window. Nothing promotes itself.
06 of 06
Depth in the work
Teams that already know the job. Not a blank canvas. Agent teams structured like the ones a growth leader already runs, each with a chief who owns a number and specialists who know the tools, the proof and the work.
- Marketing, led by MIRADRIS on organic and AI visibility, NIA on paid, MOXA on social, HOOK on creative strategy, Echo on reputation.
- Sales, led by SERANORA on outbound research and activation, DEXA on deal execution, RIVA on revenue operations, CORA on CRM conversion.
- Content, led by CREARUPA for images, FOLIO for long-form, Alpha for video, Kinetiq for motion design.
- Customer experienceConvexa, conversations that resolve and convert.
Anatomy of an agent
A configuration, not a program. Every Nagent agent is built from ten declared parts. Change one and exactly one thing changes; nothing else moves.
Technical architecture
The control plane over everything, and the room on top. People and agents meet in the team rooms. Beneath them, the agent teams; beneath those, the platform: one organisation context, orchestration, composable agents, skills, tools, models and memory, with the control plane enforcing every dimension and writing every action to the record.
Every model, every tool
The model is a setting. The tools are a grant. Each agent runs on the model that fits its work, routed through one layer that falls back when a key is exhausted and prices every call against the budget. Outside systems are reached through connectors granted one key at a time.
Deployment options
Your data where you need it. The same platform in every case. Three ways to run Nagent, one control plane across all of them. Every option keeps client data in the client’s tenant, scopes every role, and writes every action to an audit record the client can read.
Do you support open-weight models?
Yes, for enterprise deployment. In a private VPC or on premise, an agent’s model can be an open-weight model served inside your perimeter, routed through the same layer, priced against the same budget, held to the same ladder. Nagent cloud routes to frontier models through Nagent’s keys.
Enterprise and security
Enterprise controls from day one. Client data stays in the client’s tenant. Every workspace isolated, every role scoped, every agent action written to an audit log the client can read. Deploy in Nagent’s cloud, your private cloud or on premise.
- Certification
- ISO/IEC 27001:2022 certified; SOC 2 Type II programme underway
- Data protection
- Built for the DPDP Act, with GDPR-aligned handling
- Deployment
- Cloud hosted, private VPC or on premise
- Record
- Append-only audit of every human and agent action, exportable
Evaluate it
Score us on the six criteria, beside everyone else. The comparison hub scores every platform on the same six things this page is built around. Read how to run the evaluation, pick a bounded pilot, and see what each plan includes.
Frequently asked questions
What is an AI coworker in Nagent?
An AI coworker is an agent with a job, a manager and a record. It works in the same workspace as your people, takes work from MIRA, the AI chief of staff, and carries it through to pipeline and customers rather than stopping at a draft.
How does an agent earn autonomy on the platform?
Every agent starts at L2 Assist, drafting while each change waits for a person. As it builds a record of approved work and risky calls handed up, the platform recommends a promotion that a person signs off. An agent that slips loses the freedom it had.
What does the control plane show?
For every agent, its triggers, model, memory, skills, tools, budget, autonomy rung and the surfaces it works on, each with an owner and a history. Every action it takes is written to an append-only audit record your team can read and export.
Which teams does Nagent run?
Marketing, sales and customer experience, as agent teams that share one memory of your brand, your buyers and your results. Work moves between them, so content that brings a buyer to the site becomes an account NORA researches and a first message a person approves.
Where does the Nagent platform run?
In Nagent's cloud, or deployed in your own private cloud or on premise for Enterprise, so customer data stays inside your perimeter. Tenant isolation, role-based access and an append-only audit record come with every deployment.
Do you support open-weight models?
Yes, for enterprise deployment. In a private VPC or on premise, an agent's model can be an open-weight model served inside your perimeter, routed through the same layer, priced against the same budget and held to the same autonomy ladder. Nagent cloud routes to frontier models through Nagent's keys.
Go deeper
The documentation, the research and the reads.
