Skip to content

Packages

See it working on your data before you sign anything.

Four packages that put AI agents to work inside a mid-size company. Each one starts the same way: we build a working demo of the fix for your most expensive bottleneck, and you judge the result, not a proposal.

How we start

The usual route is one to three months of readiness workshops before anyone builds anything, and you pay for the workshops. We skip to the build.

  1. Step 1

    We find the problem

    From what your business already shows the world, and one short call, we pick the bottleneck that costs you the most.

  2. Step 2

    We build the demo on your data

    A working version of the fix, running on your tickets, your documents or your workload. No slides.

  3. Step 3

    You decide

    You have seen it work on your own data, so the decision rests on a result. If you say yes, the demo becomes the first piece of the production system.

Hermes agent workforce

Agents your staff message in Microsoft Teams, each limited to what that person is allowed to do.

The problem, as you would say it
“My people wait on IT, an assistant or the one colleague who knows where things are before they can get an answer or get something done.”

Example

Is the staging VPN up? Restart it if not.

It was down. I restarted it and it has been healthy for two minutes. You can do this because you are in the IT lead role.

The same request from a new hire gets the status only.

What gets installed

  • Hermes agents in your Microsoft Teams, one per function or one for the whole company
  • Roles and permissions, so the same agent does more for an IT lead than for a new hire
  • A shared knowledge base built from your documents, tickets and runbooks
  • Actions wired to your systems: infrastructure checks, deployments, mailbox and calendar work

What it replaces

The internal help queue, the “who do I ask” thread, and a personal chatbot that only knows what one person told it.

First deliverable: a working demo

An agent in your Teams, scoped to one role, answering from your own documents and completing one real task.

  • Microsoft Teams
  • Roles and permissions
  • Open-source agent runtime

embry0 software factory

An issue goes in, a reviewed pull request comes out, with evidence that the change works.

The problem, as you would say it
“Our backlog grows faster than the team can clear it, and I can't hire my way out of it.”

Example

Reject expired invite links

  • Issue read and planned
  • Change written in a sandbox
  • Review passed on the second round
  • 4 of 4 acceptance criteria checked in the browser

Pull request ready for your approval

What gets installed

  • embry0 on your own servers, connected to GitHub, GitLab or Linear
  • Triage, developer, reviewer and QA agents, each job in its own locked-down sandbox
  • Browser-verified QA: the agent boots your app and checks the acceptance criteria
  • Approvals and questions answered from the dashboard, Telegram or a GitHub comment

What it replaces

Small fixes and routine tickets queueing behind feature work, and pull requests nobody has time to review properly.

First deliverable: a working demo

One issue from your real backlog, returned as a reviewed pull request with the QA screenshots attached.

Read the source on GitHub
  • Open source, MIT
  • Self-hosted
  • Human approval

Revenue and operations agents

The work between a request and a signed deal, done by agents your reps supervise.

The problem, as you would say it
“My reps and buyers spend the day quoting, chasing replies and hunting for suppliers. The selling and the deciding get whatever time is left.”

Example

  1. 09:02Request arrives by email
  2. 09:04Priced from history and today's market
  3. 09:06Quote sent, three suppliers contacted
  4. 11:40Second follow-up, best offer accepted
  5. 11:52Contract out for signature

A rep approves the quote and the contract.

What gets installed

  • Quoting agents that price a request from your history and current market data
  • Outbound agents that contact carriers, suppliers or prospects and follow up until they answer
  • Contract push-through, from accepted quote to signed paperwork
  • Pricing models trained on your own data, and sourcing loops that find and compare components

What it replaces

Copying between inbox, spreadsheet and system of record, and follow-ups that depend on someone remembering.

First deliverable: a working demo

One workflow, a quote or a sourcing run, carried from request to result on your own data.

  • Quoting
  • Outreach
  • Supply chain

Token bill and model routing

The same work for less: every request sent to the cheapest model that can do it well.

The problem, as you would say it
“We bought the licences, everyone uses them, and the bill keeps climbing. Now we're capping usage instead of getting more from it.”

Example

Before: every request to one large model

After: each request to the cheapest model that does it well

  • Simple work, fast open model
  • Routine work, mid-size model
  • Hard reasoning, large model

Proportions are illustrative. Yours come from the audit.

What gets installed

  • An audit of where your model spend goes, by team, tool and task
  • A router that sends simple work to fast open models and hard work to large ones
  • Open models served from GPUs you own or rent, when the numbers favour it
  • Budget caps and a spend report your finance lead can read

What it replaces

One frontier model for every task, usage caps as the only cost control, and invoices nobody can explain.

First deliverable: a working demo

Your own recent workload replayed through the router, with cost and quality side by side.

  • Spend audit
  • Open models
  • Owned GPUs

Every package can run where your data already lives.

Some firms cannot send client files, patient records or controlled technical data to a public AI service. All four packages are built to run on your premises or in your own cloud account.

  • Your data stays inside your walls

    Agents, knowledge base and logs run on your servers or in your own cloud account. Nothing has to leave for the system to work.

  • Open models on GPUs you control

    When a hosted model is not an option, the same agents run on open models served from hardware you own or rent.

  • Built for firms that answer to a regulator

    Law, healthcare, defense suppliers and finance: every action is logged, every agent has a permission boundary, and a person approves what matters.

Tell me the bottleneck. I'll show you the fix.

A 30-minute call is enough to pick the problem. The next thing you see from me is the demo.

[email protected]