AgentRoom is the compliance layer between each person's assistant and your company's systems: it controls who sees what and how much is used, and records everything that touches your data. What your teams build is shared across the organization and can be offered to your customers.
Works with the assistants you already use. It doesn't replace them. Plans from USD 199 a month, paying for what you use: from USD 17 per user a month. Every company starts with a 14-day or 100,000 ART trial.
Everyone keeps the assistant they already use. AgentRoom sits between that assistant and the company's systems: it manages, supervises and audits how data is used, and controls what each user does with it.
The one each person already uses on their computer. The conversation is paid by the license they already have.
Identity, permissions, usage, approvals, data protection and a record of everything that goes through.
ERP, CRM, databases, APIs and files. Nobody reaches them without going through AgentRoom.
«Everything that touches the company's data goes through AgentRoom and is recorded.»
What we sell isn't a gateway or a number of connectors: it's the layer that lets you use any AI assistant with regulated data, and prove it.
Regulator-ready reports from the audit trail: processing activities, impact-assessment inputs, access, deletions and breaches. Templates per law in each country, starting with Chile's Law 21.719; the CMF and ASFI ones are added once their legal review is done. An obligations calendar and temporary access for your auditor. AgentRoom provides the evidence; compliance remains your company's.
Which assistants, connectors, MCP servers and skills your company uses, with alerts on anything unauthorized. It sees AgentRoom's traffic, each assistant's compliance APIs and the network logs of the tool you already use (Cloudflare, Zscaler or your firewall); for remote work, through those tools' client or an always-on VPN. AgentRoom installs nothing on devices. Personal accounts and traffic outside the company network aren't visible, and we say so.
Certified by Junngla for what companies in the region use: bank statements, SII, Defontana, Softland and Bsale, and more per country.
When an AI agent asks your company for something, a payment for example, you need to know who it is, on whose behalf it acts and what it may do. AgentRoom Trust Network is part of AgentRoom and answers that for you: the more institutions join, the safer the network is for all of them.
Each agent is registered with its identity, the person or company it acts for, its mandates and its limits.
Before each sensitive action, the network confirms the mandate is valid, the way a card is authorized today. Risky actions still wait for a person's approval.
If an agent misbehaves at one institution, the others know instantly and can block it across the network.
When an agent gets something wrong, the evidence comes from AgentRoom's audit trail.
Banks and issuers, merchants and fintechs, and agent providers. Enterprise and Corporate are network members; Equipo and Starter check whether an agent is trustworthy.
In ART, like the rest of AgentRoom: every registration, check, authorization and lookup uses ART. No membership fee.
Nobody has direct access to the ERP, CRM or databases. To reach the data you go through AgentRoom, and it's recorded.
The assistant connects with each person's account. Every query is tied to someone and their role.
In the assistants' business plans, the admin sets which connections are allowed and leaves AgentRoom as the only one for corporate data.
If the company has its assistant provider's audit logs turned on, AgentRoom adds them to its own.
What we don't promise: AgentRoom doesn't see what someone does inside their assistant outside AgentRoom. It sees, controls and records everything that touches the company's data.
Before touching your data, AgentRoom checks who is asking and what they may see. What gets through is billed in ART, by what each query costs.
What a team builds to work better can become a service for your customers, with the same pieces and a different policy.
An analyst builds a skill that reconciles the bank statement with the ERP. Their team uses it the same day.
Your customers check their account statement from their own AI assistant, and each one sees only their own.
Everything goes through a single door between AI and the company. That's where the rules apply, for any assistant.
Permissions by role, team and data. A shared dashboard shows each person only what their role allows.
One budget for the whole company, with alerts before the limit and a breakdown of where it went. In Enterprise and Corporate, also per person and per team.
Each connector and skill is enabled by role. What isn't approved doesn't show up.
Anything that writes, sends or pays asks a person to confirm before it runs, in one click from the console, email, Slack or Teams. If nobody approves in time, it doesn't run.
What's marked confidential only goes to models under a no-retention agreement. Account numbers and personal data are masked before leaving.
Who asked what, with which data and which tool ran. A record nobody can edit or delete, with an external integrity seal every day.
Everything it takes for AI to work with the company's data without losing control.
Governs information sources and development repositories, and knows each person's permissions.
Authenticates people and your customers' assistants with Google, Microsoft or your company email. SSO and automatic provisioning, as add-ons.
Roles, teams and custom roles. Who sees which source, row and column.
ART per query, person, team and company, with budgets, caps and alerts.
Whatever writes, sends or pays waits for a person to confirm.
Masks personal data and accounts, and decides which models each piece of data may go to.
A record nobody can edit or delete, with a daily integrity seal, exportable to your security system.
Skills, dashboards and automations, with their sharing levels and reviews.
Publishing, review and certification of pieces, isolated execution and payouts to publishers.
An MCP server, an API and a web widget under your brand, in every plan. It says an AI is answering, hands off to a person if you turn that on, and has a cap against abuse.
Sends each task to the right model, across Anthropic, OpenAI, Gemini and open models. In Enterprise the company chooses which models are used; in Corporate it adds its own.
A single console, showing what each role can see and do. Everything the console does can also be done from your assistant.
What someone with the same permissions already asked is answered instantly, marked CH with its age. Anyone can ask for the live data.
A vault with rotation. Neither users nor connector publishers ever see the keys. In Corporate, also with your own secrets manager or Infisical.
Reports for the regulator from the audit trail, with templates per regulation and country, as PDF with an integrity seal.
Which assistants, connectors, MCP servers and skills the company uses, also from network logs, with alerts on anything unauthorized.
Coming soon. A trust network for AI agents across institutions: agent registry, mandate authorization, shared fraud signals and disputes with evidence.
The Data Manager makes what the company already has ready for AI, in two scopes and under the same access, usage and audit rules: information, for chat assistants, and development, for coding assistants.
Your APIs connect from their spec and become tools. The MCP servers you already run sit behind the same rules.
Read-only queries, filtered by row and column per role. We recommend connecting a replica so production isn't loaded.
PDF, Excel and text, uploaded or synced from Drive or SharePoint. They're read, indexed for search and citation, and inherit their source's permissions.
GitHub and similar, for coding assistants. Repos, issues and pull requests by role; push, merge and deploy with approval. Production credentials never reach the developer.
What goes through APIs, MCP and databases is queried on the spot. If someone with the same permissions asked for the same thing recently, it's answered instantly from the cache, marked CH with its age ("CH, 5 minutes ago"), and anyone can ask for the live data. Files and their index are stored, encrypted with each company's key.
Understands what the person wants, knows which sources fit and builds the query.
Cedar decides yes or no by role, source and data, and records it. Access is never decided by AI.
The first time a source is needed, the agent discovers it and builds the route.
While nothing changes, queries go through the channel without AI: they're only checked against the rules and recorded.
Only when the source changes, permissions change or the channel fails.
Every skill, dashboard or connector moves up a level after a review. Nothing reaches further than someone approved.
Only its creator sees it
No review
Their team sees it
Team lead approves
The whole company sees it
Security and data-owner review
Your customers use it
External, legal and pricing review
Other companies see it
AgentRoom certification
Connectors, skills and templates ready to use. At launch, every piece is made by Junngla, like the local connectors for banks, SII, Defontana, Softland and Bsale; later, specialized companies publish their own and sell them to every organization on AgentRoom.
If you use less than the minimum, you pay the minimum. If you use more, you pay for what you use. The bigger the plan, the cheaper each ART.
From USD 17 per user a month, with 20 people and typical use on Starter.
The same as Starter for a small team, signed up on your own in minutes.
Start with a 14-day or 100,000 ART trial, whichever comes first.
Connect your team's AI to your tools without losing control.
Start with a 14-day or 100,000 ART trial, whichever comes first.
The whole company working with AI on its data, with rules, budget and audit trail.
Start with a 14-day or 100,000 ART trial, whichever comes first.
All of AgentRoom inside your infrastructure. No data leaves without your permission.
Over USD 5,000 a month? Let's talk.
Start with a 14-day or 100,000 ART trial, whichever comes first.
AgentRoom is in pilot. The first companies come in with our team alongside them, at these launch prices.
Each plan includes everything in the one before. What's highlighted is what you gain by moving up. Equipo has everything in Starter, with up to 10 seats and self-service sign-up.
| StarterUSD 349 · USD 0.42 / 1.000 ART | EnterpriseUSD 799 · USD 0.30 / 1.000 ART | CorporateUSD 2,199 · USD 0.26 / 1.000 ART | |
|---|---|---|---|
| People and access | |||
| Seats | Up to 20 | Up to 50 | Unlimited |
| Sign-in | Google, Microsoft or company email | Adds SSO and automatic provisioning, as add-ons | SSO and automatic provisioning included |
| Countries | One country | Several countries, with each one's rules | Several countries, with each one's rules |
| Remote session sign-out | ✓ | ✓ | ✓ |
| User roles | Admin and operator | Admin, auditor, data owner, team lead and user, plus custom roles | Admin, auditor, data owner, team lead and user, plus custom roles |
| Delegation and quarterly access review | — | Yes | ✓ |
| Usage control | |||
| Usage in view | Per person and company, with month-end projection | Also by unit, source, cost center and model | Also by unit, source, cost center and model |
| What counts as ART | Everything that goes through AgentRoom, including its AI's tokens | Same | Same |
| Usage cap | Company level | Configurable per person, team and company | Configurable per person, team and company |
| Dashboard and reports | Standard | Custom | Custom |
| Human approval | The admin approves risky actions, via console, email, Slack or Teams | Your company defines what gets approved, by whom, with a backup | Your company defines what gets approved, by whom, with a backup |
| Audit trail | 90 days, viewable in the console, with a daily integrity seal | One year, exportable, with a read-only auditor | Sent to your security system |
| Data | |||
| Data Manager | Information and development: APIs, MCP, databases, uploaded or synced files and repositories | Same as Starter | Same, inside your infrastructure |
| Sensitive data protection | Rule-based detection (national IDs, accounts, cards, emails and phones) and reversible tokens | Adds a model for free text and your own data types | That model runs locally; restricted data never leaves |
| Data classification | Four labels, with automatic classification | Adds your own labels and access requests that expire | Adds classification with a local model and integration with your tools |
| Connectors | From the catalog, including the local ones certified by Junngla | Catalog and your own | Catalog, your own and custom |
| Where data lives | AgentRoom cloud, isolated per company | AgentRoom cloud, isolated per company | Your infrastructure |
| Encryption keys | One per company | One per company | Yours |
| Credentials for your systems | AgentRoom vault, with rotation | AgentRoom vault, with rotation | Your own secrets manager or Infisical |
| Regulatory evidence | Basic report for the country's data protection law, obligations calendar and auditor access | Every template for each country and scheduled delivery | Same as Enterprise |
| Unauthorized AI detector | Assistants, connectors, MCP and skills in use, also from network logs, with alerts | Plus a per-team view for each lead | Same as Enterprise |
| AgentRoom Trust Network (coming soon) | Trust check on an agent | Member: agent and mandate registry, network fraud alerts and disputes | Same as Enterprise |
| Internal systems with no internet access | — | — | Optional local agent |
| Models | |||
| Assistants | Any MCP-compatible assistant, chat or coding | Adds those connected by key | Adds local models |
| Models for what goes through AgentRoom | Automatic routing | The company chooses which models are used | Plus its own models |
| Share and serve outside | |||
| Artifact library | User level | User, team, organization and customers | User, team, organization and customers |
| Channel to your customers | MCP server, API and web widget under your brand | Same | Same, without the AgentRoom mention in the widget |
| Marketplace access | The whole marketplace, except Corporate-only pieces (at launch, Junngla pieces) | Same, and you publish your connectors | Everything, including Corporate-only pieces installable in your infrastructure |
| Support | |||
| Rollout | Self-service | Guided | Custom project |
| Who supports you | Self-service from your assistant | Executive team as needed | Dedicated executive and SLA |
| Audit support | — | — | Yes |
Everything your company does on AgentRoom uses ART (AgentRoom Token): queries to its data, indexed and stored files, calls from its customers and model tokens that go through the platform, including its own AI's, in every plan. At month's end they're added up and billed at your plan's rate, with the minimum as the floor.
Chat inside your assistant is still paid with the license you already have. AgentRoom charges for what goes through AgentRoom.
| Equipo | Starter | Enterprise | Corporate | |
|---|---|---|---|---|
| Price per 1,000 ART | USD 0.42 | USD 0.42 | USD 0.30 | USD 0.26 |
| Monthly minimum | USD 199 | USD 349 | USD 799 | USD 2,199 |
| ART covered by the minimum | 473,810 | 830,952 | 2,663,333 | 8,457,692 |
| If you use 1 million ART | USD 420 | USD 420 | USD 799(minimum) | USD 2,199(minimum) |
| If you use 10 million ART | USD 4,200 | USD 4,200 | USD 3,000 | USD 2,600 |
The math is always the same: the greater of your plan's minimum and the ART you used times the price per 1,000 ART.
| Identity add-ons | Equipo | Starter | Enterprise | Corporate |
|---|---|---|---|---|
| SSO with your identity provider | Not available | Not available | USD 179 a month | Included |
| Automatic provisioning (SCIM) | Not available | Not available | USD 179 a month | Included |
In Enterprise and Corporate, an annual commitment lowers the price per ART: 10% from USD 12,000 a year and 15% from USD 36,000. You still pay monthly.
Equipo and Starter have no annual commitment.
| A direct query to your data, no AI | About 0.5 ART |
|---|---|
| A query with AgentRoom's AI | 5 to 100 ART depending on the model; about 34 on average |
| One person with typical use per month (50 queries a day, 15% with AI) | About 6,100 ART in queries |
| A team of 20 people with typical use per month | About 420,000 ART: fits within the Starter minimum, USD 17 per user |
An estimate with average values, adjusted with real use. It includes the company's and each seat's base on the platform. The bill is always the greater of the minimum and the usage.
Ways to buy: directly from AgentRoom, with MONZON in Chile and another payment method outside Chile. Soon also through AWS Marketplace, to use the spend commitment your company already has with AWS.
Everything is measured and shown in dollars, and every ART figure comes with its equivalent. Only the invoice is issued in local currency when needed.
The platform guides you step by step. If you'd rather have help, a certified partner does it with you.
Which data, tools and rules exist. Three use cases are chosen.
Company sign-in, roles, budgets and the first connectors.
One person per team builds on real cases and measures the savings.
Internal catalog, training and community across the company.
Spend under control, review of what's published and new cases every month.
This is how the platform is designed. Each point is verified with your team during the pilot.
The layers work together, from the moment data leaves its source until someone tries to share it outside the company.
Every source, column or file is public, internal, confidential or restricted, and the label travels with the data.
Only the rows and columns required, and aggregated results where possible.
IDs, accounts, emails and names are replaced when the assistant doesn't need the real value.
Confidential data only goes to models under a no-retention agreement. Restricted data doesn't leave.
Whatever is shared with customers, published or sent goes through a filter that blocks it or asks for approval.
Nothing reaches customers or the marketplace without its level's legal and security review.
Every generated document carries an invisible mark with user and date.
Alerts on query volumes far above normal or bulk downloads.
The limit we don't hide: if someone copies what's on their screen, AgentRoom can't stop it. So it reduces what they see, leaves a trail of everything and deters.
A recommendation, not a requirement. The model router can apply it automatically based on how the data is classified.
| Data | Recommended model | Why |
|---|---|---|
| Restricted | Open-weights model on your infrastructure, offline | The data never leaves. Lower performance than frontier models. Fits Corporate. |
| Confidential | Frontier model through your cloud (AWS, Google Cloud or Azure) | The data stays in your region under your cloud's contract, with no retention. The best balance of security and quality. |
| Internal or public | The model provider's API directly, under a no-retention agreement | Enough for data without significant risk. |
There's no big security gap between the major providers. What weighs most with a committee is the path: going through your own cloud is easier to defend than an external API.
We help you expose your services through MCP or an official API, with your rules and your security. It's a direct channel for your customers and a concrete step toward open finance.
No. It complements it: your teams keep the assistant they prefer, and AgentRoom connects it to the company's systems under your rules.
No, and we don't promise it. AgentRoom sees everything that touches the company's data, because it's the only path to it. What someone discusses with their assistant without touching that data stays between them and their assistant.
No. What's marked confidential only goes to models under a no-retention agreement, and personal data is masked before leaving. In Corporate, nothing leaves your infrastructure.
You pay your plan's minimum. If you use more, you pay for what you use, at your plan's price per 1,000 ART.
In Enterprise and Corporate you set limits and budgets per person, team and company. In Equipo and Starter you set one cap for the whole company.
In every plan, with Google, Microsoft or your company email, passwordless and with admin approval for each new person. When connecting the assistant, a second factor is required every 30 days. In Enterprise you can add SSO and automatic provisioning; in Corporate they're included.
Yes. The development Data Manager connects repositories (GitHub and similar) to coding assistants under the company's rules: each person reads the repos, issues and pull requests their role allows, and push, merge, release and deploy go through approval. Production credentials live in the vault and never reach the developer; if someone needs to use a secret, AgentRoom injects it without showing it. The terminal and the local copy stay outside AgentRoom, and everything that goes through it is recorded.
It's cancelled. If nobody approves it in time (24 hours by default), it doesn't run. It never runs by default, and each rule can have a backup approver.
Yes. Every company starts with a 14-day or 100,000 ART trial, whichever comes first.
Yes, in Enterprise and Corporate, under a publisher agreement. You receive 70% of each sale and your piece goes through the same certification as every other.
Up to 10 people, Equipo. Up to 20, Starter. Up to 50, or if you need SSO, budgets per area or several countries, Enterprise. If your data can't leave your infrastructure, Corporate.
Reports drawn from AgentRoom's audit trail: the record of processing activities, inputs for the impact assessment, access granted and denied, deletions handled, and breaches with their reporting times, as PDF with the trail's integrity seal. A calendar warns you before each deadline, such as reporting a breach, and your external auditor reviews the evidence through temporary read-only access. Equipo and Starter include the basic report for the country's data protection law; Enterprise and Corporate, every template for each country and scheduled delivery. We start with Chile's Law 21.719; the CMF and ASFI templates are added once their legal review is done. AgentRoom provides the evidence; compliance with each rule remains your company's.
Yes. When your team grows or you need more control, you move up and everything you've built keeps working.
AgentRoom is in pilot with its first companies. We pick three use cases with you, connect the first systems and measure the result before scaling.
Talk to us