Outsource AI development
Hand us the AI project. Keep the code, the data and the decisions.
TenthMatrix designs, builds and runs AI agents and generative AI applications for companies that would rather not hire a whole AI team to get one into production.
Everything we write goes into your repository from the first commit. You can read every prompt, test and trace, and anything that sends, spends or deletes waits for a person to approve it.
Why teams outsource AI work to us
Most AI projects stall after the demo. The model call works. What's missing is everything around it: connecting real systems, keeping keys out of the code, deciding what the agent may never do, testing it, releasing it safely and watching it once it's live.
That surrounding work is what we do every day for our own products. Our agents run in production on our own release pipeline, secrets manager and edge gateway, so we've already made most of the mistakes a first agent project makes.
- One team covers the agent, the integrations and the operations.
- A written scope before any build starts, including what the agent must not do.
- Hand-over is planned from day one, so you are never locked in.
Who it's for
Product teams
You have a roadmap and engineers, but nobody who has shipped an agent. We build the AI feature inside your product and leave your team able to extend it.
Startups
You need a working agent or generative AI feature in front of users before you can justify an AI hire. We build it in your repo, so it survives your first engineering hires.
Enterprises without in-house AI engineers
You have the systems, the data and the governance. We bring the agent engineering and work inside your security reviews and change process.
Agencies and consultancies
You've won AI work and need a delivery partner. We build alongside your team, with the same code-in-your-client's-repo rules.
What you can hand us
Anything from a single agent to a full application. These are the briefs we see most often.
Agentic software
Agents that work through tickets, inboxes, CRMs and repositories, then stop for review before anything goes out.
AI agent development: Agentic softwareGenerative AI applications
Knowledge assistants over your documents, drafting tools, and copilots inside the apps your teams already use.
Generative AI development: Generative AI applicationsIntegrations with your systems
Connectors, MCP servers and APIs that let a model read and act on your CRM, ERP, ticketing system, databases and file stores.
The platform around the agent
Secrets isolation, an API gateway with a WAF, health-gated releases and monitoring. We use our own products or your existing stack.
A stalled pilot
A proof of concept that worked in the demo but not on real data. We review it, write down what's missing, then finish it or tell you to stop.
Running it after launch
Monitoring, cost reports, model upgrades and a regular review, until your team takes over or for as long as you want us.
Engagement models
Pick the shape that fits how clear the job is and how much you want to own at the end. You can move from one to another.
Fixed-scope project
A defined outcome, a written scope and an agreed price for that scope. Suits a clear job: one agent, one application or one integration.
- Written scope and acceptance tests up front
- Delivered in your repository with docs
- Change requests priced before work starts
Dedicated agent team
A small TenthMatrix team that works through your backlog of AI work month by month. Suits a company with several use cases that wants steady progress.
- Team shape agreed with you
- Fortnightly planning and demo
- Scale up or down with notice
Embedded engineers
Staff augmentation: our engineers join your team, use your tools and follow your process. Suits a team that has engineers and needs agent experience among them.
- Work in your repos, tickets and channels
- Pair with your engineers to pass on skills
- You direct the day-to-day work
Build, operate, transfer
We build the system, run it in production for an agreed period, then hand it to your team with runbooks and training. Suits a company that wants ownership without the early risk.
- Agreed run period and exit criteria
- Runbooks, dashboards and on-call notes
- Hand-over sessions with your engineers
How it works, and what you get at each stage
Four stages. Each one ends with something you keep, even if you stop there.
Discovery
We pick the job, the systems it touches and who checks the result. We write down what the agent must never do and how we'll measure it.
- Written scope and risk list
- Architecture sketch and data map
- A fixed quote for the pilot
Pilot
The smallest version that does the real job on real data, behind approval gates, measured against the scope.
- Working code in your repository
- Evaluation set and results
- A go, change or stop recommendation
Build
We harden the pilot for production: integrations, secrets, guardrails, monitoring and a release path from test to production.
- Production code, tests and CI
- Security and data-handling notes
- Runbooks and dashboards
Run or hand over
We operate it with you, with cost reports and a regular review, or hand it to your team with training.
- Monitoring and alerting
- Cost and quality reports
- Hand-over sessions and docs
Security and IP
The terms clients usually ask about first.
- You own the code and the IP we write for you. It sits in your repository from the first commit.
- We sign your NDA before you share any detail.
- Your cloud or ours: we deploy into your cloud account or on-premise servers, or run it for you on infrastructure we manage.
- Model API keys and credentials stay in your secrets store. With WSL Vault, agents get short-lived, scoped credentials instead of long-lived keys.
- You choose the model providers and where inference runs, including open-weight models on your own hardware.
- Every agent action is logged, and anything that sends, spends or deletes waits for a person.
Questions clients ask
How much does an outsourced AI project cost?
It depends on the scope, so we quote after discovery instead of guessing. Discovery ends with a fixed price for the pilot. Our pricing page lists starting points for our products and services.
How soon will we see a working agent?
The pilot is designed to do the real job on real data as early as possible. We agree the timeline in the written scope, once we know which systems are involved, rather than promising a number before we've looked.
Do we need our own AI engineers?
No. We can build and run the whole thing. If you want to own it later, choose embedded engineers or build-operate-transfer and we'll train your team as we go.
Which models do you use?
Whichever fits the job and your data rules: Claude, OpenAI models, or open-weight models such as Llama, Mistral, Qwen or Kimi hosted on your own infrastructure. We keep the model behind an interface, so you can switch later.
What happens if we want to stop?
You already have the code, docs and runbooks in your repository. We hand back access and anything else we hold, and help your team or your next supplier pick it up.
Can you deliver for our agency's client?
Yes, as your delivery partner. We agree in writing how the work is presented to your client before we start.
Tell us what you want built
Send the job, the systems it touches and who checks the result. An engineer replies within one working day with questions or a call slot.