---
title: "AI"
updated: "2026-11-16"
url: "https://www.etheon.com/services/ai"
---

# AI development services

AI development, from strategy to MLOps. We build agents, models and the integrations that put them to work in your company.

## AI development, from strategy to MLOps. We build agents, models and the integrations that put them to work in your company.

### Consulting.

Where AI pays off in your company, what it needs from your data, and what to build first.

- A ranked list of use cases
- A data-readiness check
- A first build scoped

Example ranking · an illustration of the format, not a client's use cases

1. Support answers drafted from your help centre: Clear documents, high volume
2. Sales call notes written into your CRM: Repetitive work, easy to check
3. A demand forecast for your busiest products: The data exists, the value is clear

### Generative AI.

Assistants over your documents, and AI features inside your product.

- An assistant over your documents
- Answers that cite your sources
- AI features in your product

### Agents.

Agents that work in your systems, inside limits you set and with a person where it matters.

- Tools the agent may call, each with its own permission
- Limits enforced outside the model
- Every step logged and replayable
- A person approves what you choose

At this stage: Tool calling, Guardrails, Tracing

Example run · an illustration of the format, not a client's system

|  |  |  |  |
| --- | --- | --- | --- |
| 00\.00 | request | Refund order 1042 if the policy allows it |  |
| 00\.12 | retrieve | refund-policy.md, section 3.2 | grounded |
| 00\.31 | tool | orders.lookup(1042) | delivered 9 days ago |
| 00\.44 | check | inside the 14-day window | pass |
| 00\.52 | guard | amount under the agent's limit | pass |
| 00\.53 | person | approval not needed under the limit | skipped |
| 00\.71 | tool | payments.refund(1042) | done |
| 00\.72 | log | every step saved for review | saved |

### Document and vision AI.

Reading documents, images and video with AI.

- Data pulled from forms, invoices and contracts
- Scanned pages turned into text you can search
- Images sorted, and what is in them found
- Video searched for the moments you name

### Integration.

AI connected to the systems and the sign-in you already run.

- The model that fits the task
- Single sign-on
- Cost and response time measured

### MLOps.

Every version tracked, every release reversible.

- Evaluation before every release
- Versioned releases with rollback
- Model cards

### Consulting. Where AI pays off in your company, and what to build first.

#### What it is

AI consulting decides where AI is worth building in your company before anything is built. We look at your processes and your data, rank the use cases by what each can return and what it needs, and scope the first build so its result can be measured.

#### What we do

- Rank the use cases
- Check what they need from your data
- Scope a first build

#### How we do it

1. Interview: The people who know the work, and the data behind it.
2. Rank: Each use case by what it can return and what it needs.
3. Check the data: Whether it exists, is clean and may be used.
4. Scope: The first build, and how its result will be measured.

#### What you get

- A ranked use-case list
- A data-readiness note
- A scoped first project

#### Related

- [Generative AI](https://www.etheon.com/services/ai?detail=generative-ai#generative-ai)
- [How we work](https://www.etheon.com/company/how-we-work)

### Generative AI. Assistants and AI features built on your documents and your product.

#### What it is

Generative AI is software that writes, answers and summarises from your own material. We build assistants over your documents that cite their sources, and AI features inside your product, each tested against examples you approve before it reaches your users.

#### What we do

- An assistant over your documents
- Answers that cite your sources
- AI features in your product

#### How we do it

1. Gather the sources: The documents and data the assistant may read.
2. Build and ground: Answers that cite the passage they come from.
3. Evaluate: Real questions from your team, scored before release.
4. Release in stages: First to a small group, then to everyone.

#### What you get

- The assistant and its code in your repository
- An evaluation set and its scores
- A runbook for your team

#### Stack and tools

- Model APIs
- Embeddings
- Vector search
- Evaluation sets

#### Related

- [Kuest AI, an AI tutor that teaches step by step](https://www.etheon.com/work/kuest-ai)
- [Evaluation](https://www.etheon.com/services/ai?detail=evaluation#evaluation)
- [Hosted models](https://www.etheon.com/services/ai?detail=hosted-models#hosted-models)

### Agents. Software that takes actions in your systems, inside the limits you set.

#### What it is

An agent is software that takes actions in your systems: it reads a task, calls the tools it is allowed to call and records every step. We build each agent inside a harness of permissions and limits that sit outside the model, with a person approving the steps you choose, so the agent can only do what you have set.

#### What we do

- Map the task: its steps, the systems it touches, and where a person approves.
- Build the tools the agent may call, each with its own permission.
- Set the limits on data, spend and scope, enforced outside the model.
- Log every step, so your team can replay what the agent did and why.

#### How we do it

1. Scope the task: The steps, the systems and the decisions a person keeps.
2. Build the tools: Each tool the agent may call, with its own permission.
3. Evaluate: An evaluation set from your own cases, scored before release.
4. Release in stages: First a person approves every action, then only the exceptions.
5. Watch and improve: Traces, alerts and a regular review of what it got wrong.

#### What you get

- The agent and its tools in your repository
- A permission map
- An evaluation set and its scores
- A trace of every run
- A runbook for your team

#### Stack and tools

- Model APIs
- Tool calling
- Sandboxes
- Tracing
- Evaluation sets

#### Related

- [Jeriko, a team of personal AI agents](https://www.etheon.com/work/jeriko)
- [The agent harness](https://www.etheon.com/services/ai?detail=agent-harness#agent-harness)
- [ML data](https://www.etheon.com/services/data?detail=ml-data#ml-data)

Agents for your company. Tell us the task and the systems it touches. We reply within one business day.

### Document and vision AI. Reading documents, images and video with AI.

#### What it is

Document and vision AI reads what your company receives as files: forms, invoices, contracts, scanned pages, images and video. We build systems that pull out the data you need, turn scans into text you can search, sort images by what is in them and find the moments in a video you name, each checked against examples you approve.

#### What we do

- Pull data from forms, invoices and contracts
- Turn scanned pages into text you can search
- Sort images and find what is in them
- Search video for the moments you name

#### How we do it

1. Collect examples: Real documents, images or video from your work.
2. Build the reader: A model that reads them, checked against your examples.
3. Review the doubtful: A person checks what the model is unsure of.
4. Connect: The results flow into the system that needs them.

#### Questions

**Can a person check the results?**

Yes. Anything the model is unsure of goes to a review queue, and a person decides before it moves on.

**What kinds of documents can you read?**

Forms, invoices, contracts, scanned pages, images and video. We test on examples of yours before we build.

#### What you get

- The extraction pipeline in your repository
- An evaluation set and its scores
- A review queue for your team

#### Stack and tools

- Document parsing
- Optical character recognition
- Vision models
- Evaluation sets

#### Related

- [Generative AI](https://www.etheon.com/services/ai?detail=generative-ai#generative-ai)
- [Evaluation](https://www.etheon.com/services/ai?detail=evaluation#evaluation)

### Integration. AI connected to the systems you already run.

#### What it is

Integration puts AI to work inside the systems you already run. We connect a model to your applications, your data and your single sign-on, choose the model that fits each task, and measure its cost and response time so the feature stays affordable as use grows.

#### What we do

- The model that fits the task
- Single sign-on
- Cost and response time measured

#### How we do it

1. Map the systems: Where the feature lives, and what it reads and writes.
2. Choose the model: Scored on your task for quality, cost and response time.
3. Connect: Through your APIs and your single sign-on.
4. Measure: Cost and response time watched once it is live.

#### What you get

- The integration code in your repository
- A cost and response-time report
- A runbook for your team

#### Stack and tools

- Model APIs
- API gateways
- Single sign-on
- Monitoring

#### Related

- [Custom software](https://www.etheon.com/services/software?detail=custom-software#custom-software)
- [Web apps](https://www.etheon.com/services/web?detail=web-apps#web-apps)

### MLOps. Models released, measured and kept working.

#### What it is

MLOps keeps models working after launch. We build the pipelines that train, evaluate and release a model, a registry that versions each one, monitoring that watches it in production and retraining when the data moves, with every release reversible.

#### What we do

- Pipelines
- A model registry
- Monitoring
- Retraining
- Versioned releases with rollback
- Model cards

#### How we do it

1. Build the pipeline: Training, evaluation and release, written as code.
2. Register each model: Every version kept with its data and its scores.
3. Release with a rollback: A new version replaces the old only once it passes.
4. Monitor: Quality and drift watched in production.
5. Retrain: When the data moves, through the same gates as before.

#### Questions

**Can we roll back a model?**

Yes. Every release is versioned, so the previous model can return when a new one does not hold up.

**How do you know a model has drifted?**

Monitoring compares live inputs and results with the test set, and alerts your team when they move apart.

#### What you get

- The pipelines in your repository
- A model registry
- Monitoring and alerts
- A model card for each release

#### Stack and tools

- Pipeline orchestration
- Model registries
- Experiment tracking
- Monitoring
- Evaluation sets

#### Related

- [Evaluation](https://www.etheon.com/services/ai?detail=evaluation#evaluation)
- [ML data](https://www.etheon.com/services/data?detail=ml-data#ml-data)

## Inside an agent: the harness.

The model is one part; the harness around it decides what the agent may do.

- Model: Reads the task and decides the next step.
- Tools: The actions the agent may call, each one named and limited.
- Permissions: What each tool may touch, enforced outside the model.
- Context: The documents and data the agent may read for this task.
- Sandbox: Where the agent runs, apart from the systems it must not reach.
- Approval: The steps a person approves before they happen.
- Trace: Every step written to a log your team can replay.

### The agent harness. Everything around the model that decides what an agent may do.

#### What it is

An agent harness is the software around a model that decides what the agent may do: the tools it may call, the permissions on each, the context it may read, where it runs, the steps a person approves and the log of every step. We build the harness first and the agent inside it, so its limits hold whatever the model answers.

#### What we do

- Name every tool the agent may call
- Set each tool's permissions outside the model
- Run the agent in a sandbox, apart from your systems
- Put a person before the steps you choose
- Write every step to a log your team can replay

#### What you get

- The harness and its tools in your repository
- A permission map
- A replayable trace of every run

#### Stack and tools

- Tool calling
- Sandboxes
- Policy enforcement
- Tracing

#### Related

- [Agents](https://www.etheon.com/services/ai?detail=agents#agents)
- [Evaluation](https://www.etheon.com/services/ai?detail=evaluation#evaluation)

## Models.

Four ways to put a model to work, and what each asks of you.

### Use a hosted model

A provider's model, used through its API.

- What you need: An account with the provider, and terms that fit your data.
- What you own: Your prompts, your evaluation set and the code around the model.
- When it fits: When a general model already does the task well enough.

### Run an open model

Open weights on infrastructure you control.

- What you need: Servers or a cloud account with suitable hardware.
- What you own: The deployment, and where your data goes.
- When it fits: When your data must stay in your own environment.

### Fine-tune a model

A base model trained further on your examples.

- What you need: Good examples of the task, and a way to score them.
- What you own: The tuned weights, where the base model's licence allows.
- When it fits: When prompting alone does not reach your quality bar.

### Train your own

A model trained on your data for one task, such as a forecast or a classifier.

- What you need: Enough labelled data for the task.
- What you own: The model and its training code.
- When it fits: When the task is narrow and the data is yours.

### Hosted models. A provider's model, used through its API, with your data's terms checked.

#### What it is

A hosted model is a provider's model you use through its API, with no servers of your own to run. We choose between providers by scoring them on your own evaluation set, check each provider's terms against your data, and measure the cost and response time of every call.

#### What we do

- Choose by your evaluation set
- Check the provider's terms
- Measure cost and response time

#### How we do it

1. Build the evaluation set: Real examples of the task, scored the same way.
2. Compare providers: Each model scored on your set.
3. Check the terms: Where your data goes, and whether it may be kept.
4. Measure: Cost and response time at your volume.

#### What you get

- A comparison of providers on your evaluation set
- The integration code in your repository

#### Stack and tools

- Model APIs
- Evaluation sets
- Prompt management
- Usage monitoring

#### Related

- [Open models](https://www.etheon.com/services/ai?detail=open-models#open-models)
- [Evaluation](https://www.etheon.com/services/ai?detail=evaluation#evaluation)

### Open models. Open weights, run where your data already lives.

#### What it is

An open model is a model whose weights you run yourself, on servers or a cloud account you control. It suits work whose data must stay in your environment. We choose the model by your evaluation set, deploy it on your infrastructure and measure its quality, cost and speed.

#### What we do

- Choose an open model by your evaluation set
- Deploy it on infrastructure you control
- Measure its quality, cost and speed

#### How we do it

1. Shortlist: Open models whose licence fits your use.
2. Score: Each on your evaluation set.
3. Deploy: On your servers or your cloud account.
4. Measure: Quality, cost and response time under load.

#### What you get

- The deployment as code in your repository
- An evaluation of the models you considered

#### Stack and tools

- Model serving
- GPU infrastructure
- Evaluation sets
- Monitoring

#### Related

- [Hosted models](https://www.etheon.com/services/ai?detail=hosted-models#hosted-models)
- [Fine-tuning](https://www.etheon.com/services/ai?detail=fine-tuning#fine-tuning)

### Fine-tuning. A base model trained further on examples of your task.

#### What it is

Fine-tuning trains a base model further on examples of your task, so it answers in your format and your domain. It is worth it when prompting alone does not reach your quality bar. We prepare the examples, train, and score the tuned model against the base model on your evaluation set.

#### What we do

- Prepare examples of your task
- Train the base model further
- Score it against the base model

#### How we do it

1. Collect examples: Good examples of the task, checked by your team.
2. Train: The base model trained further on them.
3. Score: The tuned and the base model compared on your evaluation set.
4. Deploy: Only if the tuned model does better.

#### What you get

- The tuned weights, in your name where the base model's licence allows

#### Stack and tools

- Training frameworks
- GPU infrastructure
- Evaluation sets
- Model registries

#### Related

- [Evaluation](https://www.etheon.com/services/ai?detail=evaluation#evaluation)
- [ML data](https://www.etheon.com/services/data?detail=ml-data#ml-data)

### Machine learning. A model trained on your data for one task: a forecast, a ranking, a classifier.

#### What it is

Machine learning here means a model trained on your data for one narrow task: a forecast, a ranking or a classifier. We prepare the data, train and evaluate the model, and deploy it with the pipelines that keep it current as your data changes.

#### What we do

- Prepare your data for one task
- Train and evaluate the model
- Deploy it with pipelines that keep it current

#### How we do it

1. Frame the task: What the model predicts, and how it will be judged.
2. Prepare the data: Labelled, cleaned and split for training and testing.
3. Train and evaluate: Candidates compared on data they have not seen.
4. Deploy: With the pipeline that retrains it as your data changes.

#### What you get

- The model and its training code
- An evaluation report
- A retraining pipeline

#### Stack and tools

- Training frameworks
- Feature stores
- Experiment tracking
- Model registries

#### Related

- [ML data](https://www.etheon.com/services/data?detail=ml-data#ml-data)
- [MLOps](https://www.etheon.com/services/ai?detail=mlops#mlops)

## Evaluation: release gates.

Every model passes the same checks before it reaches your users.

| Gate | What we measure | How | Threshold | Runs |
| --- | --- | --- | --- | --- |
| Task quality | Answers correct on your test set | Scored against a labelled set you approve | Set with you | Every build |
| Grounding | Every answer cites your sources | Each sentence traced to a retrieved passage | Set with you | Every build |
| Safety | Harmful and off-policy requests refused | Red-team prompts and jailbreak suites | Set with you | Before release |
| Privacy | No personal data leaves its boundary | Redaction tested on seeded records | Set with you | Every build |
| Latency | Fast enough for the task | Response time measured under load | Set with you | Before release |
| Cost | Affordable at your volume | Cost modelled at your expected volume | Set with you | Before release |
| Drift | Still right next month | Live inputs compared with the test set | Alert level set with you | In production |

### Method: Safety

A fixed suite of adversarial prompts, plus new ones written for your domain, runs against the release candidate. Any unsafe answer blocks the release and is listed in the report you receive.

### Evaluation. Every model passes the same gates before it reaches your users.

#### What it is

Evaluation is how a model earns its release. Every model passes the same gates: task quality on a test set you approve, grounding in your sources, safety against adversarial prompts, privacy on seeded records, latency under load, cost at your volume and drift once it is live. Each gate's method is written down and runs the same way every time.

#### How we do it

1. Build the test set: Labelled examples of your task, approved by you.
2. Set the thresholds: One for each gate, agreed with you before work starts.
3. Run on every build: Task quality, grounding and privacy, as the code changes.
4. Run before release: Safety, latency and cost, on the release candidate.
5. Watch in production: Live inputs compared with the test set, so drift is caught.

#### Questions

**How do you evaluate a model before launch?**

Against the release gates on this page: task quality, grounding, safety, privacy, latency, cost and drift. We agree each threshold with you before work starts, and a release candidate that misses one waits until it passes.

#### What you get

- A report of each gate's result for every release candidate

#### Related

- [How we work](https://www.etheon.com/company/how-we-work)
- [QA](https://www.etheon.com/services/software?detail=qa#qa)

## Built and running.

A product of our own Lab, and agents we built for clients.

### Etheon Delta

Our Lab

A product of the Etheon Lab.

[Explore Delta](https://www.etheon.com/research#delta)

### Jeriko

Client work

A team of personal AI agents.

What it proves: Agents that act within the permissions you set.

[Explore Jeriko](https://www.etheon.com/work/jeriko)

### Kuest AI

Client work

A tutor that teaches step by step.

What it proves: Answers grounded in your own material.

[Explore Kuest AI](https://www.etheon.com/work/kuest-ai)

## From the Lab. Research behind the way we build AI.

[Browse all research](https://www.etheon.com/research/index)

- [What the Etheon Lab works on](https://www.etheon.com/research/index/what-the-etheon-lab-works-on): The Etheon Lab studies what makes AI software safe to ship: evaluation, guardrails, observability, safe releases and governance.
- [How we test a model before release](https://www.etheon.com/research/index/how-we-test-a-model-before-release): Every model we build passes the same release gates before it reaches users: task quality, grounding, safety, privacy, latency, cost and drift.
- [The harness comes first](https://www.etheon.com/research/index/the-harness-comes-first): An agent harness is the software around a model that decides what the agent may do. We build the harness first and the agent inside it.

## Questions.

What buyers ask before an AI project.

### How do you evaluate a model before launch?

Against the release gates on this page: task quality, grounding, safety, privacy, latency, cost and drift. We agree each threshold with you before work starts, and a release candidate that misses one waits until it passes.

### Do you train models on our data?

We never train models on your data without your opt-in. Your data stays isolated from other clients.

### Who owns the code and the models?

You do. The code, the models and the accounts are in your name, and the repository is yours from the first commit.

### How do you price a project?

By engagement model. We do not publish prices.

### Which model should we use?

The one that does your task best at a cost you can carry. We score candidate models on your own evaluation set and choose by quality, cost and response time; the four routes under Models compare the options.

- [Data](https://www.etheon.com/services/data): Pipelines, platforms and BI
- [Cloud](https://www.etheon.com/services/cloud): Cloud, DevOps and SRE