True Forge is an open-source, vendor-neutral managed agent execution harness that enables users to run AI agents locally or in private clouds with any model, eliminating vendor lock-in, runtime billing, and offering full control over workflows and data. It combines the benefits of managed agents—such as session management and tool integration—with cost efficiency, privacy, and flexibility, supported by an easy-to-use TypeScript setup and a roadmap for enhanced features.
The video introduces True Forge, an open-source, vendor-neutral, MIT-licensed managed agent execution harness developed by True Foundry. Unlike proprietary managed agents such as Claude Managed Agents, OpenAI’s new agent API, or Gemini agents, True Forge allows users to run agents fully locally or in private clouds with any model, tools, or skills they choose. This flexibility eliminates runtime billing and vendor lock-in, enabling users to customize and control their agent workflows without deep coding or dependency on hosted runtimes.
The presenter categorizes agent harnesses into three types: coding harnesses designed primarily for code generation, assemble-yourself frameworks that require manual wiring and state management, and hosted runtime harnesses (managed agents) that handle session persistence, tool calls, and approvals but are typically tied to a provider’s model and infrastructure. True Forge represents an open-source version of the third type, providing the benefits of managed agents—such as streaming, sandboxing, and UI—while allowing users to run any model locally or in their cloud environment.
A key point emphasized is that the harness, not the model, often drives the cost and efficiency of running agents. True Foundry’s benchmark showed that True Forge reduced token usage by about 60% compared to Claude Managed Agents using the same model, resulting in significant cost savings. When paired with cheaper open models, the savings can be even more substantial. Additionally, True Forge addresses concerns about data privacy, compliance, and vendor lock-in by keeping all agent definitions, prompts, and session data under the user’s control.
True Forge is implemented as a TypeScript repository with a simple setup process, including a local SQLite database and optional hosted modes with PostgreSQL or Redis. It provides a chat UI primarily for testing, an HTTP API, and a TypeScript SDK, making it easy to integrate with other applications like Slack or mobile apps. Users can configure models, connectors, skills, and sandboxes, and build custom agents that intelligently use tools and APIs to perform complex workflows, demonstrated through an example fetching AI news using a cheap open model.
The video concludes by highlighting True Forge’s roadmap, which includes gateway integrations and improved sandboxing features for secure local execution. The presenter encourages viewers to star the GitHub repo and contribute feedback, noting True Foundry’s commitment to both enterprise customers and the open-source community. Overall, True Forge offers the advantages of managed agents without the downsides of vendor lock-in, high costs, or limited model choices, making it a compelling solution for those wanting full control over their AI agent infrastructure.