AUTONAR

AI & Agents

Custom AI agents, built, tested, and governed for production

We build agents for specific jobs inside your organization, like triaging requests, assisting caseworkers, and reviewing documents against rules. Then we prove they work on your real cases and put the controls in place to run them: who owns each agent, what it can touch, and what needs a person's approval.

Start with a reviewSee how it works

Led by an Army veteran. Built by engineers who have shipped production systems for NASA and Cigna, and at Amazon Web Services (AWS), Bloomberg, and Capital One.

Request queueToday
Benefits application A-1042
Fields extracted, 2 sources attached
Ready for review
Records request R-311
Missing: signed release form
Exception
Address change C-778
Allowed to run automatically
Done
Field office policy question
Draft answer, 3 citations
Ready for review
Every step recorded in the audit log
Illustration with synthetic data, not a client system.

The agent lifecycle

Building the agent is the first third of the job

Most agents stall between a good demo and a system people rely on. We cover the whole lifecycle: build it for one job, prove it on your real work, and govern it once it's live.

BUILDEVALUATEGOVERNNot ready: back to build1ScopeOne job, one office,success criteria agreed2BuildWorkflow, tools,and permissions3EvaluateTested on yourreal cases4Launch gateMeets the criteria,or it doesn't ship5OperateApprovals, audit log,and monitoring6ImproveCorrections becomenew test casesEvery change and correction is re-tested before it goes liveAcross every stage:a named owner, least-privilege access, approval rules, an audit log, and an entry in your agent inventory
01BuildAgents for a specific job, wired into the tools your staff already use. 02EvaluateAgreed success criteria, a test set from your cases, and a launch gate. 03GovernAn owner, permissions, approval rules, and an audit trail for every agent.

01BUILD

Agents for a specific job, not a general-purpose chatbot

We start from one job your staff does every day and build an agent that does the repetitive part, inside the systems they already use. Three examples of the kind of work that fits:

Request triage and routing

Reads incoming requests, pulls out the details, sends each one to the right team, and attaches a draft response for review.

Email or formClassify and extractRight queue, draft attached

Caseworker assist

Answers staff questions from policy and case history with citations, and prepares the next document in the case.

Staff questionSearch policy and caseCited answer

Rules and document review

Checks documents or regulations against requirements and flags conflicts, gaps, and outdated language for a person to decide.

Document setCheck against rulesFlagged findings

Every agent is a defined workflow

You decide what runs on its own, what waits for a reviewer, and when it stops. Here is how a typical intake agent handles each case. Anything incomplete goes to a person instead of being guessed.

Document arrivesForm, email, or uploadExtract and checkFields, sources, rulesMissing infoAllowed actionException queueA person completes itNeeds approvalReviewer sees draft + sourcesRuns automaticallyWithin its permissionsReviewer decidesApprove, or reject with a reasonApprovedRecord updatedIn your existing systemCompleted, re-checkedAudit logEvery input, decision, approval, rejection, and change, on every path

A thin, explainable stack

Model providers' own SDKs and managed retrieval where it fits. No heavy frameworks a security reviewer has to untangle.

Explicit orchestration

Each step is a defined state that can pause for approval, resume, and write to the audit log.

Least-privilege tools

An agent gets only the systems and actions its job needs, and nothing it can't explain.

Models your environment allows

Approved commercial APIs, AWS GovCloud, or models hosted entirely on your own hardware.

02EVALUATE

Measured on your real work before anyone relies on it

We agree on what "working" means before we build, then test against it. You get a report you can take to your leadership and your security reviewers, including the results that aren't good yet.

Success criteria agreed up front

Accuracy, review effort, handling time, and cost per item, with targets set together.

A test set from your cases

Real examples, including the hard ones and the edge cases your staff already know about.

Adversarial testing

Prompt injection, attempts to reach data the agent shouldn't, and requests outside its job.

Reviewers see the evidence

Approvals show sources and reasoning, not just a confident draft. Fluent answers can persuade people even when they're wrong.

Re-tested on every change

No update ships if it makes a passing case fail. Corrections from live use become new test cases.

Evaluation report · Request triage agentv1.3 · 600 test cases
CriterionTargetResult
Routed to the right team95% or more96.8%Pass
Required fields extracted98% or more98.9%Pass
Reviewer edits per item1.0 or fewer1.4Fix
Cost per item5 cents or less3 centsPass
Prompt-injection attempts0 succeed0 of 40Pass
Recommendation: fix before launch. Drafts need too many edits. 23 failure cases attached.
Sample report with synthetic results, showing the format you receive.

03GOVERN

Know every agent you run, what it can touch, and who answers for it

Once agents are live, the questions change: who owns this one, what can it reach, when did we last test it, and how do we stop it. Governance answers those questions on any given day, not just at launch.

Agent inventory3 agents · 2 live
AgentOwnerCan accessNeeds approval forLast evaluatedStatus
Request triageOperations leadRequest inbox, routing tableAny reply sent outside the officeNov 14Live
Policy assistantBenefits managerPolicy library, read-onlyNothing; it answers staff questions onlyNov 10Live
Records reviewRecords officerCase files, read-onlyAny change to a recordOct 30Paused
Illustration with synthetic data, not a client system.

Agent inventory

Every agent with its owner, purpose, access, and evaluation history. It also feeds the AI use-case reporting agencies already do.

Permissions

Least-privilege access per tool and per system, reviewed whenever the agent's job changes.

Approval rules

Set per action, not per agent: what runs on its own, what needs a reviewer, and who that reviewer is.

Audit log

Every input, decision, approval, and change, kept in your environment and ready for review.

Monitoring

Quality, cost, and error rates tracked against the launch criteria. The owner is alerted when they drift.

Stop and roll back

Pause any agent at once, see what it already did, and restore the last approved version.

Aligned to the four functions of the NIST AI Risk Management Framework: Govern, Map, Measure, and Manage. Security and evaluation

Handover

Your team can run it without us

If the builder is the only one who can fix an agent, you don't really own it. Handover is planned from the start, and it isn't finished until your team has proven they can handle a failure on their own.

1

First weeks live, together

We run it alongside your team and keep an issues log. Everything we learn goes into the handover package.

2

Technical walkthrough

For the people who run the infrastructure: architecture, runbook, deployment, and the evaluation harness.

3

Operator session

For the staff who use it every day. They try to break it while we watch, and we fix what they find.

4

Recovery drill

Your team handles a failed run with us unavailable: notice it, pause new actions, check what already happened, and restore the last good version with their own access.

YOU KEEP

  • The code and deployment scripts
  • The test set and evaluation harness
  • The runbook and architecture documents
  • Inventory entries for every agent
  • The issues log from launch

Prefer that we keep running it? Ongoing support is available, with the same documentation either way.

Starting point

Start with a two-week review of one job

The Production Readiness Review covers controls, evaluation, running cost, and adoption for one workflow, and ends with a go, fix, or stop recommendation. It's fixed price, and the findings are yours whether or not you hire us to build.

Step 1: two weeks

Production Readiness Review

Step 2: you have a pilot

Pilot to production

Step 2: starting fresh

Production build

How the review works

FAQ

Common questions

Why hire you instead of using a tool we already own?

Often you shouldn't, and we'll tell you. Custom work makes sense when a workflow crosses systems your tools don't connect, needs approval rules or audit records they don't support, or has to run somewhere they can't, such as GovCloud or a network with no internet.

What do we get at the end of the review?

A map of the workflow, a test set built from your real cases with measured results, an estimate of running cost, and a written go, fix, or stop recommendation with a fixed-price plan for the next step.

What do you need from us?

One workflow, a sample of real documents or requests, and a few hours a week with the people who do the work. System access only as far as the review requires.

Who does the work?

Senior engineers who have built production systems for NASA and Cigna, and at Amazon Web Services (AWS), Bloomberg, and Capital One. The people who scope your project are the people who build it.

Do you use frameworks like LangChain?

Sparingly. We call model providers directly, use managed retrieval where it fits, and write the orchestration ourselves so every step is visible and easy to review. Fewer dependencies also means less for your security team to approve and maintain.

Will our data be used to train AI models?

No. With private models, your data never leaves your environment. With commercial APIs, we use enterprise terms that exclude training on your data, and only where your policies allow.

Can it run without an internet connection?

Yes. Open-weight models can run entirely on your own hardware, including on networks with no internet access.

Also from Autonar

Software & Cloud

Applications, APIs, modernization, and AWS / GovCloud.

Explore

Computer Vision

Safety alerts from existing cameras, processed on-site.

Explore

Contact

Tell us what you're working on

Send a short note: what the system does today, what isn't working, and any deadlines. An engineer will reply within two business days.

contact@autonar.ai
Email us