About Milita and the product
The product.
What is Milita?
A Salesforce assurance colleague from Release Assurance. You tell her which business processes must never break. She proves them on her own in clean cloud sessions and tells you when a result changes.
Is this Salesforce regression testing?
That is the category, yes. What is different is what gets checked. A script proves that clicking finished. Milita proves the business result held: that the 25% quote actually entered Finance Review and that sending it before approval was actually refused. You describe the process in your own words, so there is no test suite to write or maintain.
Is Milita just an AI chatbot?
She uses AI to understand what you tell her, to propose what to prove, and to explain what she found. Whether a Promise held comes from a real run in a clean session, not from a model's opinion.
Does she replay what I showed her?
No. She turns it into a Promise: the people, the steps, and the results that must hold. Proofs run from the Promise with fresh test data. A demonstration teaches her; it proves nothing.
What kinds of processes can she protect?
Anything with a clear business result: approvals, routing, record access and visibility, quotes, orders, renewals, case escalation, and handoffs between people.
Does she work on custom screens and managed packages?
She learns from the real screens and reads the real metadata, so custom and managed-package screens are exactly what she is built for.
What does she need from my team?
Someone who knows the process, a Salesforce sandbox, and a little time to describe the work and confirm what she understood. She asks about business meaning, never about automation.
Do I need to install anything?
No. Milita works in a browser in our cloud, which you watch from the conversation. Nothing runs on your computer.
Does Milita replace my testing team?
No. She protects the processes you choose and brings evidence. Your team still decides what matters and fixes what changes.
Do I still need Apex tests?
Yes. Salesforce requires them to deploy, and they catch a class of defect Milita never will. Apex tests prove your code is right; Milita proves the business process still works across real screens, permissions, and packages.
Can she replace our manual regression checklist?
For the processes you give her, she does that work continuously rather than once per release. She will not cover what you have not described, so most teams start with the handful that would hurt most if they broke quietly.
How does she know what to test?
Two ways. You tell her in a sentence. Or she reads your sandbox's structure, objects, rules, flows, approvals, and proposes Promises for each, as a positive and a negative case, with the person each needs. You keep or skip each proposal.
How long until I see value?
The free plan keeps 3 Promises. Connect a sandbox and you can watch Milita prove one real process in your first session. We do not publish a minute count until we have measured it.
What happens when a Proof fails?
You get a Change: what is different, when it last worked, when it first failed, and what was recorded in between. Infrastructure trouble is reported as infrastructure trouble. An uncertain result is inconclusive. Only a new Proof marks recovery.
Why doesn't she tell me what caused it?
Because she cannot know, and a guessed cause is worse than none. She lists the events between the two proofs, a deployment among them, and leaves the conclusion to you.
Can my pipeline use her?
Yes. The pipeline starts a Proof of the selected Promises through a scoped API key, waits, and proceeds only on a pass. Exit zero only on passed; anything else holds the release. Release gate.
What is a kept Promise?
One Milita has proved on her own at least once and is actively checking. Until that first clean Proof it is a draft.
What if she can't find a control?
She asks you what it's called or where it is, in the conversation, and builds the step from your answer. She remembers your answer as an anchor that expires if the screen changes too much.
About security
Security.
Is it safe to connect our Salesforce org?
Milita connects only to sandboxes, never production, and only with a non-admin Salesforce user you control. Admin-level users are refused at connect and before every run. You can revoke access in Setup at any time. Security.
Why sandboxes only, with no exception?
Because a proof that touches live data is a liability, not a proof. A full or partial sandbox that mirrors your configuration gives the most faithful result with none of the risk.
Does she need admin access?
No, and she will not use it. She works with the same access as the person the process belongs to. For reading org structure she can use a separate, API-only reader that cannot sign in or act, and every read is logged for you.
Do you store our Salesforce password?
Never. Access is certificate-based, tokens are short-lived and in memory only, and no refresh token is kept.
Where does our data live?
Google Cloud. Customer data is separated in the database itself. Evidence is private, served through your own sign-in, and deleted on the schedule your plan sets, from 7 days on Free to one year on Enterprise.
What does the AI see?
Names, labels, values and outcomes on the pages and the structure she reads, bounded and fenced as data. Never Salesforce credentials, API keys, run evidence, formula text verbatim, or another customer's rows. Requests go to Google Vertex AI and are not used to train models.
Are you SOC 2 certified?
No, and we will not say so until an independent audit is complete. We are preparing for a SOC 2 Type 1 audit expected to cover security and confidentiality, with a written system description, adopted policies, and dated control evidence.
Can we get a DPA and complete a security questionnaire?
Yes. The DPA is available for review on request, and we support questionnaires and architecture reviews directly. Email security@releaseassurance.com.
Can you see our computer?
No. Nothing of hers runs on your computer. She works in a browser in our cloud, signed in to your sandbox.
What happens if Salesforce asks for a step-up challenge?
The run stops and reports a clear failure. Milita never attempts to work around an identity challenge.
About pricing
Pricing.
Is the free plan really free?
Yes. 3 Promises and 20 Proofs a month, no card, no time limit, and fair use of 60 runs and 150 Milita model calls a month. We would rather you proved a few processes forever than abandoned a trial.
What counts as a Proof?
One run of a kept Promise, up to 10 minutes of its steps; each further 10 minutes counts as another Proof. A debug, the checks Milita runs while you set up or activate a Promise, and a Proof that fails because of us are never counted. Your account shows the Proofs used this month, and the minutes they ran.
What happens if I use my Proofs up?
On a paid plan your Proofs keep running: each further Proof is 10¢ on Starter and 5¢ on Pro and Business, charged to your card after the month ends. On Free, new Proofs pause until your next month. Your existing Proofs and evidence stay exactly where they are. Pricing has the detail.
Can I change plan later?
Yes, in both directions. Until card checkout opens, email hello@releaseassurance.com and we change it for you. Billing & refunds says how that month is settled.
Do you work with consultancies?
Yes. Talk to us if you manage several client orgs.
Do you offer invoicing?
Monthly on Starter, Pro and Business. Card checkout opens soon: until then, email hello@releaseassurance.com and we move your workspace to a paid plan. Enterprise is annual and invoiced under your agreement, with purchase orders supported.
What does she cost compared with a test automation tool?
The free plan keeps 3 Promises forever with no card. Starter is $150 a month. Most scripted tools price per seat or per test and carry the ongoing cost of whoever maintains the scripts. Pricing.
About AI and determinism
AI and determinism.
If AI is involved, how can the result be deterministic?
The model interprets, proposes, and explains. The verdict comes from a real run of a validated plan against your real sandbox, graded by typed assertions. The same plan against the same org produces the same result, and no model opinion can turn a failure into a pass.
Can the AI write code that runs against my org?
No. Everything Milita builds is a plan of known primitives, data, never code. The model cannot emit browser code, and a missing capability becomes a named platform gap rather than an escape hatch.
Why is a separate checker better than an AI testing its own work?
An AI that builds the change and grades the check has a conflict of interest: it will loosen the assertion, work around the failure, or declare success on the wrong evidence. Milita is independent of whatever wrote your configuration and grades the real org state.
Which model does Milita use?
Google models served through Vertex AI. Which model answers is an operator setting with one-click rollback, and a new model is admitted only after it clears an evaluation corpus for authoring fidelity, questions asked, false assumptions and injection resistance.
Can someone trick Milita through my org's data?
Tool results and org data reach the model fenced as data, never as instructions, and an adversarial corpus is run against the served engine. Consequential actions are never executed on the model's decision; you accept them from a card.
About the company
The company.
Who is behind Milita?
Release Assurance Inc., founded by Adrian Rudzikas, a Salesforce architect with twenty years in the ecosystem, 18 times certified. Based in Alberta, Canada. About.
What does AI-first mean at Release Assurance?
AI is how we build and run the company, not a side project or a feature added to the product. AI coding agents write the platform under the founder's direction, and Milita, the product, is an AI colleague. Being AI-first does not mean trusting AI: an agent cannot approve its own work, a person approves every change, and a model never decides whether a Promise held. How we work.
How can one person run a production platform?
Because the company is AI-first. AI coding agents he directs do the building, under a change process built to audit standards: every change through review, an automated gate, and a merge queue; production only through a release he approves; agents with their own identities and no access to production data. More than 2,200 pull requests in six months, every one on the record.
Are you a Salesforce partner?
Not yet. Release Assurance is an independent company. We will say so the day it is official.
Do you publish customer names?
Not yet. We publish customer names, quotes and numbers only with permission, and we have not asked anyone to lend theirs.
How do I reach you?
hello@releaseassurance.com for anything, security@releaseassurance.com for security questions and vulnerability reports. The founder replies within 24 hours.