Agents, people and rules in one panel
From the panel you create agents, decide what each one can do, give each person access and review every run. An assistant proposes the configuration; you decide.
Read the transcript
You create an agent for your team in a system under your control.
You give it the code: a repository or a ZIP file.
In seconds, the pre-analysis recognizes your system and what the agent can do.
You see what it will read, what stays out and what to review first.
The analysis reads the code and turns it into knowledge.
And it checks the result against what you confirmed.
The summary shows what it understood about your system, with the evidence.
Use cases, rules, screens and connections, all in one catalog.
You review each executable use case: its steps, its data and its status.
Once it has everything it needs, you publish it. Only what’s published runs.
You choose what the agent can do and what waits for approval.
And in the system rules you decide who approves: someone else, not whoever launched the task.
You connect your systems and the operator’s computer, with the Reelease Agent.
You invite an operator: they only see the chat and the agents you choose.
They get the invitation and start working with the agent.
From your systems to an agent under control, ready for your team.
How the pieces fit
Every task crosses the same checkpoints. Your people ask and approve; Reelease builds, seals and records; the Reelease Agent runs only what was sealed, and your systems receive calls pinned to sealed steps.
Swipe to see the full diagram.
- 1
Ask. An operator asks in plain language.
- 2
Propose. The model suggests a plan.
- 3
Seal. Rules validate each step; the plan is signed.
- 4
Approve. Changes wait for Two-Key Approval.
- 5
Execute. The agent verifies the signature and runs.
- 6
Call. Your system receives the sealed call.
- 7
Record. Each step lands in the run record.
Checking, sealing, approving and recording are done by Reelease, without the model.
The Reelease service runs on Google Cloud in Madrid, with its encrypted database in Frankfurt. Connection credentials are stored encrypted in Reelease and used only for the sealed call.
You decide what each agent does
Each agent’s rules are limits it cannot skip: what it can do, what waits for a person and how far it can go. Only an admin can change them.
Everything starts with no permissions
A new agent can’t touch any system until an admin turns on what it needs. And what the system’s rules forbid, no agent can allow.
- What it can do
- An admin turns on each agent’s capabilities one by one: your systems, email, documents, the browser or the operator’s computer, the last one always with their consent.
- What gets approved, and by whom
- By default, every change to your systems waits for someone other than the person who launched the task. Each step can have its own approver and deadline. The approver sees the exact values of each change.
- Pre-authorized changes
- Routine tasks with fixed data don’t have to wait: an admin can pre-authorize them and record why. Pre-authorization never covers what the agent decides, and the run record shows it.
- Limits
- By default, at most 30 steps and 10 changes per plan, and every action is checked again while it runs.
Every rule change is recorded with who, when and why. Risky changes are confirmed first, and you can go back to an earlier version.
See everything you controlEveryone, with just the access they need
You invite each person and decide which access type they get, which role they have and which agents they work with. Admins govern from the panel; operators work from the chat, only with their agents.
- Platform · web panel
Admin
- Invites people and manages their roles and agents
- Sets each agent’s rules and who approves
- Decides who can approve in an emergency
- Resumes a suspended agent
- Platform · web panel
Member
- Sees and configures only the agents assigned to them
- Prepares knowledge and executable use cases
- Assigns who approves each step
- Can suspend an agent if something looks wrong
- Operator · chat only
Operator
- Works from the chat with their assigned agents
- Asks, co-writes and runs tasks from their own computer
- Uses their own work accounts; nobody else sees their credentials
- Follows their runs and handles the approvals assigned to them
- Invitations with access already set
By email, with the access type, role and agents. They expire after 7 days and can be resent or canceled.
- Agent-by-agent access
Members and operators only see the agents you assign them; admins see them all.
- Who approves each step
By default, the approver is someone other than the person who started the run, and each step can have its own approver.
- Offboarding without losing the history
Removing someone revokes their agents, connected accounts and computers, and keeps their history.
Every agent starts with no permissions
Access per person and per agent, the approvals you define, execution in an isolated folder with signed plans only and a record of every run.
- Lucía Prieto launches the task
- Lucía Prieto reviews and approves the plan
- Andrés Molina approves change RTDC
- Andrés Molina approves change RTRY
- Run completed
Frequently asked questions
Who uses it?
In the web panel (“Platform”), admins govern the system and members configure their agents. In the chat (“Operator”), each person asks, drafts and runs tasks with their agents, within the rules.
Who decides what each agent can do?
Only an admin changes the rules or installs agent templates, and every change is versioned with who, when and why. Members prepare their agents’ knowledge and executable use cases.
How do we remove someone?
An admin removes them from the system: they lose their agents and connected accounts at once, and their computers can no longer run anything. Their runs, approvals and records are kept.
What does the configuration assistant do?
It proposes drafts of use cases and configuration on the panel’s screens. It applies nothing on its own: whoever has permission reviews the proposal and decides.
See it on your own system
We’ll show you Reelease on one of your cases: what it understands, the plan an agent would propose and what you’d sign off on.