An open-source AI agent stack runs on your machine. Ares adds the limits it leaves to you.
A workflow tool, an agent framework and a local model server will run in your building at a low cost. Ares runs on the same kind of machine, and adds an identity for each agent, inspection, approvals and one record of the workflow.
A decade in defence and aerospace. In government procurement since 2016. The same technology since 2019, validated in US DoD and Canadian DND programs.
A workflow tool, an agent framework and a local model server, all open source, fit on one small machine. A capable engineer or agency can assemble them in little time.
What an open-source stack is good for
It is cheap to start, it runs offline, and you can read every line. Local models keep the data on your hardware. The workflow tools in that world connect systems well, and many have a step where a person approves before the flow continues. A large community writes connectors and answers questions. For experiments, internal tools and tasks a person checks, it is a sound choice.
Where it stops
Assembled from separate projects, the stack has no single owner for the parts that limit what an agent may do.
- Agent identity. Agents usually share the credentials stored in the workflow tool. Each connection holds whatever access it was given, for every run.
- Inspection. A document goes to the model as it arrives. A hidden instruction in a supplier email reaches the agent, unless someone adds a filter and keeps it current.
- The record. Each project keeps its own log, in its own format. Joining them into one record of the workflow, refusals included, that an outsider can check, is work your team does.
- Licences. Read each one, for limits on commercial use, hosting for others or redistribution. They matter when a partner builds the stack for you or you offer the workflow to a customer.
- Updates. Each part has its own release cycle and its own security notices. Someone has to watch all of them and patch the machine.
Side by side.
One small machine, offline
- An open-source stack you assemble
- Fits, and runs with no outside link.
- Ares
- Fits. A whole site runs on one small machine, with no outside link.
Licence cost
- An open-source stack you assemble
- Low or none.
- Ares
- A licence or subscription.
Agent identity
- An open-source stack you assemble
- Shared credentials in the workflow tool.
- Ares
- Each agent signs in as itself and gets a credential for one job.
Human approval
- An open-source stack you assemble
- An approval step in some workflow tools.
- Ares
- A named person approves the steps you choose, and the decision is on the record.
Inspection
- An open-source stack you assemble
- Added and maintained by your team.
- Ares
- Documents inspected on the way in, answers checked on the way out.
The record
- An open-source stack you assemble
- One log per project.
- Ares
- One record of the workflow, refusals included, checked offline with an open-source verifier.
Models
- An open-source stack you assemble
- Local models of your choice.
- Ares
- Your choice, local models included.
Who answers for it
- An open-source stack you assemble
- Your team, or the firm that assembled it.
- Ares
- We do, for the platform and for any workflow we build with you.
By category. Projects differ and change often. Check the version you would run.
How they fit together
Ares does not ask you to give up open source. The verifier that checks the record is open source, so anyone can check a record without us. The open source page lists it. Agents written with open-source frameworks connect over A2A, MCP or our SDK and run under Ares without a rewrite. Your local models stay.
What changes is around the agent. Each agent signs in as itself. Each document is inspected before it is read. A person takes the sensitive steps. The workflow leaves one record. The deployment page shows Ares on one small machine in your building, connected or not.
They do a different job. They can work together.
What Ares does not do
- It is a licensed product, not free software. The verifier that checks its record is open source.
- Local models differ from the largest cloud models. We say which runs where.
- Filters cover stated patterns and languages.
- A flow built in a workflow tool may need to be rebuilt as an Ares workflow. We say which before a pilot.
Questions and answers
Tell us about one workflow.
Tell us what you would assemble and the task it would do. We will say what Ares adds, what it does not, and what the first step costs.
A senior member of the team replies within two business days.