UNSUAI LAB

AI AGENT IMPLEMENTATION

Review a process

CUSTOM IMPLEMENTATION / THREE MONTHS

That process can keep moving without someone chasing it all day.

Show us how the process works today, which information it needs and where it gets stuck. We build an agent that reads approved sources, prepares the work and uses the tools agreed in the proposal.

Your team knows what the agent may do, what still needs approval and what happens when information is missing. We choose Codex, Hermes or another foundation after we understand the work.

  • 03 MONTHSto build and test
  • ARCHITECTUREchosen for the process
  • HUMANreview for important decisions
  • USD 5,000starting point
IMPLEMENTATION / 01STATUS: ASSESSMENT

PROCESS FIRST, TECHNOLOGY SECOND

We understand the process before choosing the technology.

INPUT
The data, file or message that starts the work.
SOURCES
Approved sources and accounts.
WORK
The repeatable part of the agreed scope.
WHO DECIDES
The person who reviews the result.
PROCESS → ARCHITECTURE → TEST → HANDOFF01
IF INFORMATION IS MISSING

The workflow stops and alerts the responsible person.

YOUR TEAM KEEPS

Context · agents · tests · documentation

The implementation connects part of the work
to the tools your company already uses.

The Applied AI program helps a team document a task and test it in tools such as ChatGPT or Gemini. A person still runs the process step by step.

Implementation starts when an agent needs to read approved sources, move information between tools, run on a schedule or prepare actions for approval.

The course is not a prerequisite. If the process is clear and has a responsible owner, we can begin with implementation.

An agent needs context, tools,
permissions and supervision.

Inside a company, an agent needs operating context, tools, permissions, run records and a clear way to stop.

01

What the agent needs

07 PARTS · OPEN
  1. 01A model that can reason about the case.
  2. 02Files and context that preserve rules and examples.
  3. 03Tools for reading and carrying out work.
  4. 04Permissions that limit what the agent can access or change.
  5. 05Memory so the work does not start from zero.
  6. 06Schedules for recurring work.
  7. 07A person who reviews results and handles exceptions.
02

What your company knows

05 PARTS · OPEN
  1. 01The person who knows the process.
  2. 02Examples that worked and examples that did not.
  3. 03Approved sources, accounts and permissions.
  4. 04The exceptions that appear in real work.
  5. 05A human standard for reviewing the result.

The agent's responsibility determines the right foundation.

We review where the data lives, which accounts the company uses, what the agent must do, where it can run and who will operate it.

01 / 03THE OPEN FOUNDATION

For a company that wants model choice and more control over its infrastructure.

Hermes is an open-source agent harness created by Nous Research. It can work with different models and providers, run on company-owned or remote infrastructure, and connect to tools through skills and MCP. It can also preserve memory, delegate work and run scheduled tasks.

It can be a strong fit when the company needs model choice or more control over infrastructure.
  • More than one model
  • Persistent operation
  • Agreed infrastructure
  • Less vendor dependence
KNOWN LIMITOpen source does not mean free or secure by default.

We review the whole company.

We automate work with explainable rules, approved sources and someone who can review it. Some processes should remain human-run.

01 / 12AUTOMATION POSSIBILITY

Leadership and operations

The agent can prepare a daily brief, gather updates, maintain a decision log, follow up on open items and flag work that remains blocked.

HERMESRecurring work and alerts.

CODEXFiles, boards, reports and internal tools.

THIS MAP SHOWS THE OPTIONS. THE PROPOSAL DEFINES THE SCOPE.

Show us where the work keeps piling up.

That is where we decide which agent your company needs. We look for a clear input, approved sources, rules people can explain and someone who can review the result.

01 / 04WE KNOW WHAT IT RECEIVES

The email, file, form or message that starts the process.

We identify who sends it, how it arrives and which information must be complete before the agent can act.

  • Source
  • Format
  • Permission
PROCESS AGREEDINPUT DEFINED

We turn operating experience into rules an agent can follow.

Your company brings the people, examples, sources, permissions and exceptions. UNSU builds the architecture, environment, context, connections, tests and documentation.

WHAT YOUR COMPANY BRINGSJUDGMENT
  • The person who knows the process.
  • Good and bad examples.
  • Approved sources and accounts.
  • A human review standard.
WHAT UNSU BUILDSAGENTS
  • Hermes, Codex or both.
  • Local, AWS or another agreed environment.
  • Skills, memory and connections.
  • Tests, alerts and handoff.

EVERYTHING RUNS WITH THE ACCOUNTS, PERMISSIONS AND COSTS LISTED IN THE PROPOSAL

We learn the process, connect the tools, then try to break the workflow.

MONTH 0101

Understand and choose

The person who knows the work walks us through it. We review inputs, sources, rules, exceptions, risks and owners. Then we choose Hermes, Codex or a justified combination.

QUEDA LISTO

A process map covering sources, owners, risks and technical scope.

MONTH 0202

Build and connect

We configure the environment, prepare the context and connect the tools included in the proposal. The first cases run under supervision.

QUEDA LISTO

Configured agents, environment, connections and initial workflows.

MONTH 0303

Test and hand off

We run incomplete cases, different formats and real exceptions. We adjust rules, permissions and alerts. The team learns how to operate the system and receives the documentation.

QUEDA LISTO

Tests, controls, documentation, training and handoff.

The agent handles the agreed work.
Your company keeps the important decisions.

APPROVED SOURCES ONACTIONS LOGGED ONPERMISSIONS BY TOOL ONHUMAN REVIEW ONIF INFORMATION IS MISSING STOP

Each workflow defines what the agent may read, what it may change, what it only prepares and what requires approval. Financial, legal, employment and other high-risk decisions keep a responsible human owner.

You receive an operating system, not a demo.

You receive the agreed scope and a way to operate it. If a source changes or a new case appears, the team knows what to review before running the workflow again.

  1. 01

    Automation map

    Reviewed processes, priorities, sources, owners, risks and opportunities for later phases.

  2. 02

    Architecture decision

    Hermes, Codex or both, with the reason for that decision and the dependencies the company needs.

  3. 03

    Configured environment

    AWS, local infrastructure or another agreed environment, with access and costs documented.

  4. 04

    Agents and workflows

    Context, instructions, skills, memory, scheduled tasks, connections and actions explicitly included in the proposal.

  5. 05

    Tests and controls

    Normal cases, exceptions, logs, alerts, permissions and human review points.

  6. 06

    Runbook and training

    A manual for operating, reviewing and modifying the implementation, followed by handoff to the responsible team.

OWNERSHIPYour data, accounts and outputs remain yours. The proposal identifies the configuration and code you receive, the open-source components and any external services the implementation requires.

FromUSD 5,000

Includes assessment, architecture, configuration of Hermes, Codex or both, the contracted workflows, testing, documentation and handoff. Final price depends on the number of processes, tools, sources, integrations and security requirements.

Discuss my implementation

A few things companies usually ask before we build.

01Do Hermes and Codex do the same thing?

They overlap on some tasks, but their foundations differ. Hermes supports an open, multimodel operation across different infrastructure. Codex begins in the OpenAI ecosystem and works directly with software, files, skills, plugins and connected services.

02Do I need to choose one before we talk?

No. We first review the process, existing accounts, permissions and where the system should run. Choosing the architecture is part of our work.

03Can you use both?

Yes, when each one has a clear responsibility and the handoff can be tested and reviewed. Using two agents without a reason adds cost and maintenance.

04Will you automate every department?

We review the areas that matter to you and build a map. The proposal names the processes included in the three-month scope. The starting price does not cover automating an entire company.

05Do I need to take the course first?

No. The course helps a team learn the method and can make adoption easier. Implementation can begin directly when there is a clear process, approved sources and a responsible owner.

06Is Codex included with ChatGPT?

Availability and limits depend on the plan and workspace. During the assessment, we review the company's existing access and any additional services, connectors or usage it may need.

07Does open source mean there are no monthly costs?

No. Hermes does not charge a license for its open-source code, but infrastructure, models, APIs and maintenance may cost money. The proposal lists those costs.

08What about sensitive information?

We work with approved sources and access. We review special security or compliance requirements before closing scope because they can change the architecture, permissions and price.

09What happens after three months?

Your team receives the contracted implementation and documentation. Ongoing support, monitoring or additional processes are scoped separately if needed.

Bring us one process that should no longer depend on someone chasing it.

On the first call, we look at how the process starts, which information it reads, which tools it touches and who recognizes a good result. Then we decide whether it should be automated and which architecture fits.

Review a process in my company