Introduction
Triform manages the lifecycle of AI agents which autonomously gather information, make decisions and execute sequences of actions to achieve their goals.
As the designer of an agent, you define the agent’s purpose along with its domain-specific context, such as necessary tools, API:s and data sources. For example, a customer service agent might require information about your organization in order to give helpful answers to your customers.
We provide you with a platform which helps you build, evaluate and deploy your agent without requiring the help of backend and AI engineers:
Build agents as durable Python workflows with your favorite LLMs, databases and libraries, or let our agent-building agent do it for you.
Evaluate quantitative and qualitative performance metrics using our evaluation framework.
Iterate rapidly with introspection: live execution traces provide intermediate results, execution times and evaluation scorecards for your workflows.
Deploy your agents on our purpose-built serverless architecture with the click of a button, providing millisecond latency cold-starts and strong function-level isolation.
Monitor your agents continuously and configure alerts to trigger on deviations in trends (WIP).
Share your finest creations with the rest of the community, helping other creators just like you while earning Triform credits in return (WIP).
Why Triform?
Building reliable agents spans multiple distinct disciplines: frontend, backend and AI. Adopting AI comes with large costs in building and maintaining knowledge and infrastructure.
Triform provides an integrated platform incorporating best-practices and guardrails, letting you focus on designing the agents rather than everything around them. Chances are you won't even have to build your agent: we provide a community-driven library of free components and an agent-building agent ready to assist you.
LLMs struggle to build anything but shorter snippets of code, even LLMs supporting larger context windows find prioritizing within them challenging. Agentic solutions like Lovable also struggle significantly with maintainability.
By defining an agent as a workflow, our AI builder can create the flow graph with clearly defined schemas for inputs/outputs, and delegate the implementation of the Python functions to separate builders. Each builder only considers a single Python function and its desired inputs/outputs, significantly reducing the size of the context.
This granular modularity also promotes re-use of your agents and functions within your own library, your organization or if you feel like sharing, with everyone in the community.
Demand for and legislation revolving around robust data governance and operational control is on the rise. Organizations are looking to choose their own cloud providers, or run their services in their own datacenters.
Triform was created with on-premise hosting in mind. The platform does not rely on any proprietary cloud services, and declarative deployment manifests (helm) make it easy to deploy in any modern environment supporting Kubernetes. Get in touch if you want to know more!
What does an agent look like?
Agents come in many shapes and sizes. An agent taking meeting notes might appear as a passive participant in your meetings. A trading analyst might frequently email you regarding market opportunities.
AI-centric apps can transition seamlessly between different agents, invoked via our API. Agents can in turn use other agents as tools via MCP, and so on. Many agents don't have a look, as they take on supporting roles where end-users don't interact with them directly.
Everyone can't be the star of the show, but what if you need one? Conversational agents are common, and providing a chat UI out of the box is on our roadmap. For more complex use-cases, you can integrate Triform agents with your existing UI or build custom interfaces tailored to your needs.