An AI agent framework developed and released by Scale AI. It lets you assemble agents at whatever level of complexity you need, from a simple chatbot to a fully autonomous system. Its defining trait is a design built around long-running asynchronous agents: durable workflows via Temporal, plus Redis-backed streaming and persistent state management, are included from the start. The same setup carries you from local development with Docker Compose all the way to cloud operation on Kubernetes. It was open-sourced under the Apache 2.0 license on November 12, 2025, and is also offered as the agent infrastructure layer of the Scale GenAI Platform (SGP).
Key Features
- Five agent levels: The CLI scaffolds templates matching your goal, from a simple chatbot (L1) to a fully autonomous agent (L5). You don’t have to build a complex setup up front — you can move up a level when the need actually arises
- Durable workflows through Temporal: Designed for processes that take minutes to hours and may fail partway through. Intermediate progress isn’t lost, so resumption and retries work
- Async architecture and streaming: Messaging through Redis streams model output token by token while state is persisted on the server side
- Developer UI and trace visualization: A local web UI lets you test agent responses, debug, and inspect execution traces. Auto-reload on code changes is supported
- Multi-agent configurations: The Agent Developer Kit (ADK) lets sub-agents communicate with each other so you can build systems with divided roles
- Kubernetes-native and cloud-agnostic: Runs under Docker Compose or in a Docker-free local mode, and deploys as-is to your own Kubernetes cluster. No lock-in to a particular cloud vendor
Pricing
| Plan | Price | Key features |
|---|---|---|
| Open source edition | Free (Apache 2.0) | All functionality including the Agent Server, developer UI, Python SDK, and CLI. Local development and self-hosting supported. Support comes from the GitHub community |
| Enterprise Edition | Contact sales | Everything above plus Scale-managed GitOps/CI-CD, AgentOps (hosting, version control, tracing, evaluation), model inference and knowledge bases, SSO/SAML, and uptime SLAs |
Pricing reflects information as of August 2026. Enterprise Edition pricing is not published and is handled on an inquiry basis. Check the official documentation for the latest details.
Pros & Cons
✅ Pros
- State persistence, resumption, and tracing — everything long-running async agents need — are built in, sparing you from implementing them yourself
- Local development (Docker Compose) and production (Kubernetes) run on the same configuration, reducing incidents caused by environment drift
- Apache 2.0 open source, so you can read the source, self-host, and avoid vendor lock-in
- CLI scaffolding lowers the barrier to getting started with agent development
- You can migrate to the Enterprise Edition later to satisfy organizational requirements such as SSO and SLAs
⚠️ Cons
- Requires Python 3.12 or later and an understanding of surrounding components like Redis and Temporal. It’s easily overkill if you only want a simple chatbot
- Production operation demands Kubernetes expertise, which is a heavy burden for small teams without infrastructure staff
- Released in November 2025, so community knowledge and non-English material are still limited
- Enterprise Edition pricing is undisclosed, making it hard to estimate adoption cost in advance
- Not a no-code agent builder — writing code is a prerequisite
Comparison with Similar Services
| Criteria | Scale Agentex | LangGraph | CrewAI | Google Agent Development Kit |
|---|---|---|---|---|
| Provider | Scale AI | LangChain | CrewAI | |
| License | Apache 2.0 | Open source | Open source | Apache 2.0 |
| Strength | Production operation of long-running async agents | Flow control via graph structures | Role-based multi-agent systems | Integration with the Google ecosystem |
| Durable workflows | Standard, via Temporal integration | Checkpointing mechanism | Depends on external implementation | Depends on external implementation |
| Deployment | Kubernetes-native, cloud-agnostic | Managed offering available | Managed offering available | Centered on Google Cloud integration |
| Commercial edition | Enterprise Edition (contact sales) | Yes | Yes | Billed through Google Cloud |
Who Is It For
- Development teams that need to run long-lived agents in production, such as research or automation tasks that take tens of minutes to several hours
- Engineers who want to design monitoring, tracing, and version management in from the start rather than stopping at a prototype
- Companies that want to run agents on their own Kubernetes infrastructure without being tied to a specific cloud
- Technology evaluators who want to inspect the source while deciding whether to adopt it
- Teams that expect organizational requirements like SSO and SLAs down the line and want a migration path secured
Summary
Scale Agentex is a framework that confronts the operational problems — asynchrony, long-running execution, and state persistence — that leave many AI agent projects stuck at “the demo works but production breaks.” Its Temporal integration and Kubernetes-native design suit serious production use, but the learning curve and operational load are far from trivial. Start by running the open-source edition locally, watch the traces in the developer UI, and judge whether this level of complexity matches your use case.