Give your AI the access it needs. Without handing it the keys.
Organize instructions, contexts, skills, credentials and scripts in one place. The AI connects through MCP, runs scripts on controlled runners and uses each credential without ever accessing its value.
Works with any MCP-compatible client — Claude, Cursor, VS Code and more. One endpoint per project: mcp.iaimigo.com/mcp/p/your-project
Your AI already works. What's missing is a proper place for it to work.
Every team repeats the same instructions in every chat, pastes credentials into the conversation and loses the scripts the AI wrote yesterday.
Scattered instructions
Project rules live in old messages, loose files and the head of whoever started it. Every new chat starts from zero.
Credentials in the chat
For the AI to reach a database or an API, someone pastes the password into the conversation. It ends up in logs, histories and the model's context.
Throwaway scripts
The AI writes a useful script, runs it once and it's gone. Nobody knows what ran, with which access, or how to repeat it.
Three steps between your team and an AI that works safely
Nothing to install in your model. IAImigo sits between the AI and your infrastructure.
Organize
Create projects and register instructions, contexts, skills, credentials and scripts. Decide what applies organization-wide and what belongs to each project — open to the whole project or restricted to the people and groups you choose.
Connect
Add the project's MCP connector to your AI client. The login is yours (OAuth 2.1), and the AI receives the instructions plus the skill and context indexes as soon as the conversation starts.
Run
When the AI needs to act, it requests a run — with a title and a purpose. If a project rule requires it, the run waits for approval; then a controlled runner injects the credentials, executes the script in isolation, stores the result and records everything in the audit trail.
Five things every AI needs — organized, versioned and access-controlled
Every item applies to the whole organization or to a single project, is open to the entire project or restricted to whoever you choose, and has tags. The AI gets exactly what it is allowed to see.
Instructions
How the AI should behave in the project: rules, style, terminology. Short and always loaded, in every conversation.
Contexts
What the AI needs to know. Large documents loaded on demand by topic, with a markdown editor and search.
Skills
How the AI gets a task done: a step-by-step guide that loads only when the topic calls for it, with the contexts, scripts and credentials it needs. Open SKILL.md format, compatible with Claude Code.
Credentials
Organization-wide, per project or personal, kept in a vault. The AI knows name and description; the value only reaches the script, inside the runner.
Scripts
An immutable history of everything that ran — each run with its title, purpose and who asked for it — plus a curated catalog of reusable scripts with versions, pins and usage counters.
The AI uses a credential without ever accessing its value. Neither does our platform have to.
Security isn't a feature of IAImigo — it's the reason it exists. Every architecture decision starts there.
- Credentials injected in the runner — the value goes from the vault straight into the script's process, encrypted end to end. It never passes through the chat or the model's context.
- Credentials with owners — every credential has owners — a person or a whole team — who replace the value when needed, without ever having to see it. Personal credentials belong only to whoever holds them and cease to exist along with the account.
- Authorization outside the code — every access is decided by centralized policies, in the context of whoever asked. Viewing, using and revealing a credential are separate permissions, and emergency access (break-glass) requires dual approval.
- Isolated execution — one ephemeral container per run, sandboxed, no network by default and resource limits. mTLS with short-lived certificates between runner and platform.
- Real audit trail — who asked and why, which script, which credentials were used (by name), on which runner, through which AI client and with what result. Distinct events for “credential used”, “credential revealed” and “value changed”.
Run wherever it makes sense: your machine, your server or our platform
One runner, four scopes: user, project, organization and platform. You set the dispatch order per organization or project, with automatic fallback to the next tier — and can switch any of them off.
On your machine
A Docker container on your computer. Ideal for reaching VPNs, SSH, internal databases and local files. Only runs what you asked for yourself.
On the project's server
A runner reserved for one project — the machine that reaches a client's legacy database, for example. It only takes work from that project, and you decide who can use it.
On your infrastructure
A runner on your servers, available to every project in the organization. Scripts run close to your systems: internal networks, VPNs and databases that aren't on the internet.
On our platform
Nothing to maintain: ephemeral containers isolated per run, a curated image with the most-used tools and explicit limits.
Each runner has labels, such as internal-network, gpu or production, and a plain-language description of what it can reach. The script states what it needs, the AI reads the descriptions, and the platform only dispatches to a runner that meets the request and that you're allowed to use.
Every AI task becomes a session your whole team can follow
The AI doesn't remember yesterday's conversation, and your teammate can't see yours. The session keeps the thread: the goal, every run and what the AI decided along the way.
Session log
Every run, every AI note and every decision goes in, in order, with no editing. In the next conversation, the AI picks up from the latest summary instead of starting from scratch.
Several people, one task
A teammate joins your session, and their AI sees what has already been done. If you both try to run something at the same time, the AI knows what's running and decides whether to wait or go ahead — nobody steps on anyone else's work.
Approval before running
Organization and project rules define what needs approval — by script type, credential, runner or time of day — and who approves, in one or more levels. With no answer in time, the run is rejected, unless you configure otherwise.
Built for companies that take access seriously
Roles and permissions
Owner, administrator, security manager, auditor, project manager, editor, credential manager, executor and reader — with granular permissions per resource and action. Revealing a credential's value is a separate permission.
Sign-in and one account per company
Sign in with your Google account or with email and password, with MFA. Work for more than one company? Each one has its own account, with its own access rules, and you switch between them with the account switcher without signing out.
Multilingual and time zones
Interface in English, Portuguese and Spanish; timestamps in each person's time zone, with the organization's default.
Usage and retention
Retention classes per script type, usage metering per project and expiration policies applied automatically in storage.
Limits and grants
Set limits per person or role: runs per day, runner minutes, allowed credentials and runners, time of day. For a one-off task, approve a budget for that session only, with an expiry date.
Isolation per customer
Database, cache, vault, storage and identity separated per organization. One company's data never mixes with another's.
A plan for every team size
We're opening access in waves. Prices will be published at launch — accounts created now get the launch conditions.
Starter
For small teams that want to stop pasting passwords into chats.
- Projects, instructions, contexts and skills
- Credentials in the platform vault
- Platform runners and runners on your machine
- Audit trail and run history
Business
For companies with several teams, projects and environments.
- Everything in Starter
- Runners on your infrastructure
- Advanced roles, approval rules and break-glass
- SSO with your identity provider
- Configurable retention and webhooks
Enterprise
For those who require isolation, data residency and compliance.
- Everything in Business
- Vault and storage in your cloud (BYO)
- Dedicated instance and region of choice
- SCIM, SLA and dedicated support
- Runner source code audit
No credit card required to create an account. Final prices and limits will be published before billing starts.
Frequently asked questions
Which AIs does IAImigo work with?
Any client that supports the MCP (Model Context Protocol): Claude (web, desktop and Code), Cursor, VS Code and others. You add the project's connector to the client, sign in and that's it — no need to change model or tool.
Can I use the skills I already have in Claude Code?
Yes. IAImigo skills use the open SKILL.md format, the same one Claude Code uses: you import the skill and it becomes available to the project, with access control and version history. It works the other way around too.
Does the AI really use a credential without accessing its value?
Yes. The AI only receives the name and description of the credentials you authorized. When a script needs one of them, the runner fetches the value from the vault and injects it directly into the script's process. The value does not pass through the chat, the model or the returned result.
Do I need to host anything?
No. You can use only the platform runners. If you'd rather have scripts run on your machine or your infrastructure — to reach internal networks, VPNs or databases that aren't on the internet — just start a runner with one command.
Where is my data stored?
Each organization has its own database, cache, vault and storage, isolated from other customers. Your organization's data stays in a single region and is never queried from any other.
What happens when someone leaves the company?
When the account is deactivated, everything it holds is paused. When it's deleted, that person's personal credentials are deleted with it. Organization and project credentials keep working, because they have owners — if that person was the only one, an administrator assigns another, without seeing the value. If a personal value is needed before deletion, there's emergency access with dual approval. Everything is recorded in the audit trail.
How much does it cost?
Plans and prices will be published at commercial launch. Creating an account is free and requires no card; those who join now get the launch conditions.
Give your AI a proper place to work.
Create your account, connect the first project and never paste a password into a chat again.