Skip to content

Home Case studies Appstrate

Creation of an open source platform to run AI agents, isolated and under control

Appstrate, product preview

Apache 2.0

open source license, free to install on your own servers

1 container

isolated per run, destroyed at the end

2026

year the platform was created

Appstrate is the open source platform to run AI agents, created by TRACTR. Each agent works in an isolated container, never sees the credentials of the services it uses, and every run is traced. TRACTR uses it every day in its own operations.

View the product

At a glance

ClientTRACTR (studio product)
EngagementTRACTR product: design, build and operation
SectorAI infrastructure
Technologies
  • TypeScript
  • Bun
  • Docker
  • PostgreSQL
  • Redis
  • MCP
  • OpenAPI

Context and challenges

The most capable coding agents run on a developer’s machine: one agent per person, with their own access, hard to share, control or audit. TRACTR created Appstrate to run them centrally: each run is isolated, access is managed by role and credentials are never exposed to the model. The platform is open and follows an open packaging standard (AFPS) that describes an agent as code.

What we did

One isolated container per run

Each run starts in a fresh Docker environment, destroyed at the end. Hardware isolation through a micro virtual machine is available as an option.

Credentials the model never sees

A proxy injects access to services (OAuth 2.0, API key…): the agent works with Gmail, Slack, ClickUp, Notion or Stripe without ever reading the token.

Agents described as code

An agent is a versioned package, with its instructions, skills and tools, following the open AFPS standard: it can be reviewed, versioned and moved from one host to another.

A platform for several teams

Organizations, roles, API keys, cron scheduling, real-time run monitoring and a documented API (OpenAPI).

Open code

The code is published under the Apache 2.0 license, and the platform installs with one command on your own servers.

Video

In pictures

  • The dashboard: recent agents, latest runs and upcoming schedules.

  • The workspace agents, published with their keywords.

  • The chat launches an agent and summarizes its result.

  • Agents can be scheduled: here, a watch every day at 8 a.m.

  • Several model providers, with an order of preference.

  • Everyone can also use their own model subscriptions.

The dashboard: recent agents, latest runs and upcoming schedules.
The dashboard: recent agents, latest runs and upcoming schedules. 1 / 6

More case studies

All our case studies

The next step

What are we moving forward together?

An idea, an existing product, a task that takes too long. Tell us where you're starting from.

Let's talk about your projectA first conversation to understand your context.