AI that does real work on your systems, by your rules
Reelease understands and documents your systems from their code. Your team publishes what each agent can run, and the agent only takes those steps, each with its own rules. By default, changes wait for another person, and everything goes into a run record.
Read the transcript
The safe way to put AI to work on critical systems and in regulated sectors.
Here, a mistake is a transfer that never goes out or a missed deadline.
Laura asks for what she needs and the agent proposes a plan.
Reelease checks the rules and the permissions…
…and signs the plan: if anyone alters it, it doesn’t run.
The lookups go ahead on their own.
The change waits for someone else: whoever starts the task can’t approve it.
María approves and the transfer goes out, on the same core as always.
Resolved within the shift, with nothing rewritten.
Every task leaves its run record: who asked, what was done and who approved it.
And you can stop an agent instantly.
On the systems you already run. No migration, no rewrite. And it keeps working when you replace them.
And how does the agent know what to do?
First, Reelease has read your system, exactly as it is.
It reads the code and the documentation and builds a map your team reviews.
And it leaves everything documented, with its sources, as PDF or Word.
Your team chooses which use cases each agent can execute.
Only what your team publishes can run. Everything else stays documented, for reference.
Every agent starts with no permissions: an admin turns on only what it needs.
Your AI, where you decide.
It runs where you decide, with the AI model you choose.
Your data is looked up, never copied. Nothing is used to train models.
For sectors that can’t improvise: the same control, with concrete cases in every sector.
Reelease, the control layer between AI and your systems.
AI for critical systems. Under control.
Reelease is the control layer between AI and your systems: its agents work on what you already run and do only what your team publishes, you decide which steps a person approves, and every run leaves a run record, the complete audit trail of the task.
One executable use case combines the core, your tools, AI steps and people, so you can improve the process too.
Once AI starts to act, control has to live inside every run.
| What it is | What it isn’t |
|---|---|
| Your system understood from its code: its map, its documentation and its use cases, with their rules, screens and data, each with its source. Your team reviews them and turns the ones it chooses into tasks for your agents. | A wiki the agent reads without anyone reviewing it, or documentation that sits in a drawer. |
| Agents that work on what you already run, from the mainframe to your email. | A migration or rewrite project, like those from integrators, IBM watsonx Code Assistant for Z or AWS Mainframe Modernization. |
| Deterministic: it runs the Sealed Plan you signed, task by task, and your rules check every step first. With Reelease’s own engine, the default for plans with a connector, a step with all its data is compiled from the Sealed Plan: the model doesn’t write the call. Otherwise the call stays pinned to the sealed step, and changes are approved on their exact values. It does what you sign off on. Nothing more. | An agent that does whatever its natural-language instructions say. You talk to Reelease the same way, but what you write becomes a plan your rules check before it runs. |
| Access per person and per agent, starting from zero. | An agent with access to your whole core and no rules. |
| No hallucinations: a chat that answers with its source, in your code or your documents, checks each citation against it and tells you when there is none. It also drafts with you. | A generic AI chat, such as ChatGPT, Claude or Gemini. |
| Changes a person approves by default, and a record of every run. | A coding copilot that proposes changes nobody controls. |
| A bot that automates screens (RPA), such as UiPath, Automation Anywhere or Blue Prism. | |
| Agentic process automation, such as Maisa or UiPath Agentic Automation. |
- Your system understood from its code: its map, its documentation and its use cases, with their rules, screens and data, each with its source. Your team reviews them and turns the ones it chooses into tasks for your agents.
- Agents that work on what you already run, from the mainframe to your email.
- Deterministic: it runs the Sealed Plan you signed, task by task, and your rules check every step first. With Reelease’s own engine, the default for plans with a connector, a step with all its data is compiled from the Sealed Plan: the model doesn’t write the call. Otherwise the call stays pinned to the sealed step, and changes are approved on their exact values. It does what you sign off on. Nothing more.
- Access per person and per agent, starting from zero.
- No hallucinations: a chat that answers with its source, in your code or your documents, checks each citation against it and tells you when there is none. It also drafts with you.
- Changes a person approves by default, and a record of every run.
- A wiki the agent reads without anyone reviewing it, or documentation that sits in a drawer.
- A migration or rewrite project, like those from integrators, IBM watsonx Code Assistant for Z or AWS Mainframe Modernization.
- An agent that does whatever its natural-language instructions say. You talk to Reelease the same way, but what you write becomes a plan your rules check before it runs.
- An agent with access to your whole core and no rules.
- A generic AI chat, such as ChatGPT, Claude or Gemini.
- A coding copilot that proposes changes nobody controls.
- A bot that automates screens (RPA), such as UiPath, Automation Anywhere or Blue Prism.
- Agentic process automation, such as Maisa or UiPath Agentic Automation.
- A stable layer between your agents and your systems: when you replace one, you change its connection and the agents carry on as before.
- Whatever the analysis finds stays a proposal until you publish it.
Brands named are examples of each category and belong to their owners. Their presence does not imply affiliation or endorsement.
AI agent glossaryAudit trail, guardrails, human-in-the-loop: what each term means and what Reelease calls it.See the glossaryCompared with what you’re already evaluating
Migrating, automating screens, an AI chat or process agents. Reelease understands your system from its code and works on the core under your control, in Europe.
| Migrate or rewriteFor example: Integrators, IBM watsonx Code Assistant for Z, AWS Mainframe Modernization | RPAFor example: UiPath, Automation Anywhere, Blue Prism | Generic AI (chat)For example: ChatGPT, Claude, Gemini | Agentic process automationFor example: Maisa, UiPath Agentic Automation | ReeleaseFrom your code and on your core, in the EU | |
|---|---|---|---|---|---|
| How it understands your system | Partly: Through an analysis at the start of the project, to rewrite it | No: From what it sees on screen | No: From what you tell it in the conversation | Partly: From the process as it is described and configured | Yes: From your code Reads the code and documentation, from Oracle Forms and Reports to your APIs, and maps your system and its use cases, with the source of every fact |
| Work on your core | No: It replaces it: everything is rewritten and moved | Partly: Through the screen, and it breaks when the screen changes | No: It doesn’t touch it: someone acts by hand afterwards | Partly: Mostly back-office processes and documents | Yes: On the core Mainframe and CICS, PL/SQL and databases inside your network. Only the executable use cases you publish act on it |
| What runs | Partly: The new code, once the project ends | Partly: A fixed script of clicks and keystrokes | No: Nothing: it only answers | Partly: The configured process, with its rules and exceptions | Yes: Deterministic The Sealed Plan Reelease builds and you sign, task by task, and your rules validate every step. With Reelease’s own engine, the default for plans with a connector, the model doesn’t write the calls. If the plan changes, it doesn’t run |
| Human approval | No: The project is approved, not each change | Partly: Added by whoever designs the bot, if they plan for it | No: The person decides what to do with the answer | Partly: Human review at the points defined when the process is configured | Yes: You approve every change With Two-Key Approval, your rules set who approves (someone else, by default). You approve the exact values; if they change, you’re asked again |
| Answers about your system | Partly: Whatever documentation the project leaves | No: It doesn’t answer questions | Partly: From what the model knows, with no source in your code | Partly: From the process knowledge it was given | Yes: No hallucinations Every fact about your system comes with its source, in your code or your documents, and each citation is checked against it; when there is none, it says so. No invented action reaches your systems |
| Proof of every run | No: The project’s documentation, not each task’s | Partly: The bot’s logs | No: A chat thread, with no shared record | Partly: A log of the process steps | Yes: A run record for every run Every step, who approved it and which account talked to each system. Exportable, and you can replay the run from it |
| Where your data lives | Partly: Wherever the project contract says | Partly: Wherever the bots are deployed | Partly: Wherever the chat service runs | Partly: In the region the vendor offers | Yes: In the EU Service in Madrid and models inside the European Union, or in your cloud or on your premises |
| When you see value | No: At the end of a long project | Partly: Soon, on screen tasks | Partly: Right away, but only as answers | Partly: Soon, on back-office processes | Yes: From the first use case With your first executable use case published, on what you already run |
| When you replace a system | No: It is the replacement: all at once | No: The bots have to be rebuilt | No: Nothing changes: it doesn’t run anything | Partly: It depends on how each process was connected | Yes: No lock-in You switch the connection; agents, approvals and run records stay the same |
How it understands your system
Reelease
Yes: From your code
Reads the code and documentation, from Oracle Forms and Reports to your APIs, and maps your system and its use cases, with the source of every fact
The alternatives
- Migrate or rewrite
Partly: Through an analysis at the start of the project, to rewrite it
- RPA
No: From what it sees on screen
- Generic AI (chat)
No: From what you tell it in the conversation
- Agentic process automation
Partly: From the process as it is described and configured
Work on your core
Reelease
Yes: On the core
Mainframe and CICS, PL/SQL and databases inside your network. Only the executable use cases you publish act on it
The alternatives
- Migrate or rewrite
No: It replaces it: everything is rewritten and moved
- RPA
Partly: Through the screen, and it breaks when the screen changes
- Generic AI (chat)
No: It doesn’t touch it: someone acts by hand afterwards
- Agentic process automation
Partly: Mostly back-office processes and documents
What runs
Reelease
Yes: Deterministic
The Sealed Plan Reelease builds and you sign, task by task, and your rules validate every step. With Reelease’s own engine, the default for plans with a connector, the model doesn’t write the calls. If the plan changes, it doesn’t run
The alternatives
- Migrate or rewrite
Partly: The new code, once the project ends
- RPA
Partly: A fixed script of clicks and keystrokes
- Generic AI (chat)
No: Nothing: it only answers
- Agentic process automation
Partly: The configured process, with its rules and exceptions
Human approval
Reelease
Yes: You approve every change
With Two-Key Approval, your rules set who approves (someone else, by default). You approve the exact values; if they change, you’re asked again
The alternatives
- Migrate or rewrite
No: The project is approved, not each change
- RPA
Partly: Added by whoever designs the bot, if they plan for it
- Generic AI (chat)
No: The person decides what to do with the answer
- Agentic process automation
Partly: Human review at the points defined when the process is configured
Answers about your system
Reelease
Yes: No hallucinations
Every fact about your system comes with its source, in your code or your documents, and each citation is checked against it; when there is none, it says so. No invented action reaches your systems
The alternatives
- Migrate or rewrite
Partly: Whatever documentation the project leaves
- RPA
No: It doesn’t answer questions
- Generic AI (chat)
Partly: From what the model knows, with no source in your code
- Agentic process automation
Partly: From the process knowledge it was given
Proof of every run
Reelease
Yes: A run record for every run
Every step, who approved it and which account talked to each system. Exportable, and you can replay the run from it
The alternatives
- Migrate or rewrite
No: The project’s documentation, not each task’s
- RPA
Partly: The bot’s logs
- Generic AI (chat)
No: A chat thread, with no shared record
- Agentic process automation
Partly: A log of the process steps
Where your data lives
Reelease
Yes: In the EU
Service in Madrid and models inside the European Union, or in your cloud or on your premises
The alternatives
- Migrate or rewrite
Partly: Wherever the project contract says
- RPA
Partly: Wherever the bots are deployed
- Generic AI (chat)
Partly: Wherever the chat service runs
- Agentic process automation
Partly: In the region the vendor offers
When you see value
Reelease
Yes: From the first use case
With your first executable use case published, on what you already run
The alternatives
- Migrate or rewrite
No: At the end of a long project
- RPA
Partly: Soon, on screen tasks
- Generic AI (chat)
Partly: Right away, but only as answers
- Agentic process automation
Partly: Soon, on back-office processes
When you replace a system
Reelease
Yes: No lock-in
You switch the connection; agents, approvals and run records stay the same
The alternatives
- Migrate or rewrite
No: It is the replacement: all at once
- RPA
No: The bots have to be rebuilt
- Generic AI (chat)
No: Nothing changes: it doesn’t run anything
- Agentic process automation
Partly: It depends on how each process was connected
- Migrate or rewrite:
- Integrators, IBM watsonx Code Assistant for Z, AWS Mainframe Modernization
- RPA:
- UiPath, Automation Anywhere, Blue Prism
- Generic AI (chat):
- ChatGPT, Claude, Gemini
- Agentic process automation:
- Maisa, UiPath Agentic Automation
Brands named are examples of each category and belong to their owners. The table compares categories, not specific products.
Frequently asked questions
What is Reelease?
The control layer between AI and your systems: a platform that understands and documents what you already have, draws out its use cases and puts AI agents to work on them. They run only what your team publishes, with the approvals you define and a record of every run.
Is it a copilot or a chatbot?
No. A copilot answers in a chat. Reelease agents act on real systems: they check their rules before starting, get human-in-the-loop approval on the steps you choose and record what they do.
Can we improve a process as well as run it?
Yes. Building on what Reelease understands about the process, you can add steps that are done by hand today (an email, a ticket, a lookup in another system), combine data from several systems in the same task and replace pieces one at a time. Each new step has its own rules and approvals and is part of the run record.
What’s the difference between AI agent governance and security?
Governance decides what agents can do and who can do what: rules, executable use cases, approvals and per-person access. Security protects the environment: each operator’s identity, least privilege by default, encrypted credentials, Sealed Plans and records.
How do we get started?
Tell us about a process or a system that matters to you. We’ll show you Reelease on a similar case and, if it fits, run a real 3-month pilot on your own systems with you.
Who is Reelease for?
Companies and public bodies with critical systems that can’t improvise: banking, insurance, the public sector, healthcare, justice, telecom, energy, industry and defense. Your systems can be decades old or recent: it works on the mainframe and on your APIs.
How does approval work?
You decide which steps wait for a person and who approves them. By default, Two-Key Approval applies, the four-eyes principle: anything that changes data waits for someone else, never the person who started the task, unless an admin allows it with a reason that goes into the run record. The approver sees the exact values of the change first, and if those values change, the approval no longer counts.
Where is my data? Is any model trained on it?
Inside the European Union. On the Reelease service, services run in Madrid and the encrypted database is in Frankfurt, and the AI models Reelease offers also run in the EU. Your databases are looked up on each task, not copied, and no model is trained on what the agent reads.
Which technologies does Reelease understand, and which can it act on?
It understands and documents the mainframe (COBOL, CICS, PL/I, JCL, IMS, VSAM, Db2 and MQ), Oracle (PL/SQL, Forms and Reports), IBM i (RPG, CL and DDS), Natural and Adabas, .NET Framework, Java, PowerBuilder, Delphi and VB6, Python, Ruby, NestJS, GraphQL and React. An agent acts through five paths: CICS transactions on the core; your web APIs (REST, SOAP, GraphQL and PL/SQL web); your Oracle, SQL Server and PostgreSQL databases through the Client Bridge; the operator's computer, for desktop applications; and each person's work account or the organization app, for Microsoft 365, Gmail, Jira, Confluence, Azure DevOps, LinkedIn, Slack and GitHub. What has no direct path, such as JCL or web front ends, it understands so you can document and renew it.
Does it work with decades-old systems?
Yes, and with modern ones too. Reelease understands the mainframe (COBOL, CICS and PL/I), IBM i, Natural, PL/SQL, Oracle Forms and Reports, .NET Framework, PowerBuilder, Delphi and VB6, as well as Java, Python, Ruby, Node, React and your APIs. Agents run what you publish, and there’s nothing to migrate first: you start on what you already run.
Does Reelease migrate my COBOL?
It isn’t an automatic migration. Reelease documents your COBOL, gives agents the executable use cases you publish and, if you decide to replace it, what it understands about the system (business rules, screens and flows) is your starting point. You replace it piece by piece, at your own pace, and everything else keeps running. The market calls this legacy modernization.
What does each run leave on record?
Its run record keeps the complete audit trail of the task: the Sealed Plan and its hash, who approved each change and when, every action on your systems with its outcome, and the rules that applied. You can export it and put it on legal hold, and it's kept even if you delete the agent. A complete sample run record is on the site for you to open.
Does it help with DORA, NIS2 and the EU AI Act?
It helps. The EU AI Act, DORA and NIS2 ask for human oversight, change control and traceability, and Reelease gives you controls and evidence for each, mapped by article: the run record supports the logging of Article 12 of the AI Act, and the approvals you define support the human oversight of Article 14. For DORA (Articles 28 to 30), you get the documentation to assess us as an ICT provider. The assessment is yours; we give you the evidence, the DPA and the subprocessor list.
How much does it cost?
You can start free, with one system and one agent. To try Reelease on your own systems with hands-on help there is the Pilot, and for production, Pro: both have published prices. Enterprise is custom, by contract.
See plans and pricingStart with a process that matters to you
We’ll show you Reelease on a case like yours and, if it fits, set up a scoped pilot.