AI apps that make it to production
We build AI products on open-source stacks and run them on your infrastructure, for startups and enterprise teams. The tools we pick come from the research behind the 875 projects we rank.
- Fixed scope, clear milestones
- Your data stays in your account
- You own the code
- Remote from Vietnam, UTC+7
Where we help most
A prototype that has to go live
Built on Lovable, Bolt or Supabase, it works in the demo and breaks with real users. We find what fails first and fix it in order of risk.
A pilot that must pass security review
The demo impressed, and now security, legal and platform teams have questions. We build the version that answers them.
Data that cannot leave your servers
You want an assistant or search over internal documents, and a third-party API is off the table. We run open models next to your data.
Engineers on coding agents
The agents write code faster than the team can review it. We set up the rules and gates so speed does not cost quality.
What we build
AI products, end to end
Assistants, agents and search over your own documents, from a first prototype to a production release with tests, monitoring and a runbook.
You get: A working app in production, its test set, dashboards and a runbook your team can follow.
Typical work
- A support assistant that answers from your help center and hands off to a person
- Search across Drive, Confluence or SharePoint that cites the page it read
- Contracts, invoices and forms turned into structured data your systems can use
- Agents that act in your CRM or ticketing tool, with a person approving each step
Private AI on your own infrastructure
Open models, an AI gateway and a vector database on servers you control, in your data center or your own AWS, Google Cloud or Azure account.
You get: The stack running in your account, as code, with access control, logs and a cost estimate per month.
Typical work
- Open models served on your GPUs or a GPU cloud you pick
- One gateway in front of every model, with keys, budgets and logs per team
- A vector database next to your data, kept in sync with its source
- A clear rule for when a hosted model is allowed and which data it may see
Stack review before you commit
An independent read of the open-source tools you plan to adopt, before your team builds on them.
You get: A written report with one pick per need, the risks of each, and a migration path if a pick goes stale.
Typical work
- License terms, including relicensing risk and what your use case needs
- Maintenance health: who maintains it, how often it ships, open security issues
- Fit with your team, your hosting and the data rules you work under
- Two or three alternatives for each pick, with the trade-offs
AI coding workflows
Coding agents set up for your engineering team, with the rules and gates that keep quality where it was.
You get: The setup in your repositories, the CI gates, and a guide for the team.
Typical work
- Shared rules and project context the agents read in every repository
- Review gates in CI: tests, lint and security checks before a person reviews
- Clear limits on what an agent may touch, such as secrets, production and migrations
- A short training session and a guide your team keeps
For enterprise teams
Built to pass your security review
What a security, legal or platform team asks first is part of the build, not a later phase.
Your data stays where you say
Models run on your servers or in your cloud account. When a hosted model is used, it is one that does not train on your data, and only on the data you approve.
Access and audit
Sign-in through your identity provider, roles per team, and a log of every model call: who asked, what was sent, what came back, what it cost.
Measured before launch
A test set built from your real questions, scored before every release, so quality is a number your team can track.
Guardrails
Personal data masked before it reaches a model, defenses against prompt injection, and a person approving any action that changes data.
Costs you can predict
A monthly estimate before we build, budgets per team in the gateway, caching, and smaller models where they do the job.
No lock-in
Open-source parts with licenses checked, each one swappable, and everything in your repositories as code.
How a project runs
Brief
Answer four short questions by email. A rough answer is fine.
Scope and quote
One call, then a written scope with milestones, a fixed price for each, and the monthly running cost.
Build
Short milestones with a demo and a test score at the end of each. The code sits in your repository from day one.
Handover
Docs, runbook, dashboards and every access are yours, plus a session with your team.
Ways to work together
Audit
Fixed price
A production-readiness or stack review with a written report. The smallest way to start.
Project
Fixed price per milestone
A defined build, from prototype to production. You pay a milestone when its demo meets the agreed test.
Ongoing support
Monthly
Monitoring, model updates and the test set rerun whenever a model or prompt changes.
Prices come with the scope, after the brief and one call.
Work you can check today
Choosing the stack is the costly mistake, so we built the research first and run it in public. Your app is built on it.
archestack.com
875 open-source projects in 2 ranked lists, rebuilt from GitHub data twice a day and served from Cloudflare's edge.
@archestack/mcp
Our MCP server on npm. Claude, Cursor and other AI apps use it to search, rank and compare the lists.
The research
License, last commit and known weak spots for every project we rank, so each pick for you comes with its risks.
Client work never changes a rank. Our independence rules
Questions teams ask
Do you work with teams outside Vietnam?
Yes. We work remotely from Vietnam (UTC+7) with teams in any region, in English.
Will you sign our NDA and fill in our security questionnaire?
Yes. We sign your NDA before the first call that needs it, and answer your vendor security questions in writing.
Can the AI run without sending our data to OpenAI or Anthropic?
Yes. We serve open models on your own servers. When a hosted model fits the job better, we say so, show what it would see, and you decide.
Who owns the code?
You do, from the first commit. It lives in your repository and runs in your accounts, with no license fee to us.
Can you work alongside our engineers?
Yes. We work in your repositories, tickets and chat, pair with your team on the hard parts, and hand over with a session for everyone who will run it.
How do you price a project?
A fixed price per milestone, agreed after the brief and one call, plus an estimate of what the system costs to run each month. Ongoing support is a monthly fee. We invoice by bank wire, Wise or PayPal.
Do you build with tools from your own lists?
Often, because the lists are where our research lives. Client work never changes a rank, including the rank of the tools we build with.
Tell us what you are building
The brief email opens with these four questions. Answer what you can.
- What are you building, and who will use it?
- What runs today (stack, models, hosting)?
- What should be live, and by when?
- Anything we must respect (data rules, budget range, team)?
Goes to hello@archestack.com. Ken replies within one business days.