Run Jev inside an Open Agent Harness

TrueForge, an open-source agent harness, simplifies the process of turning Large Language Models (LLMs) into functional agents. It handles various aspects such as model calls, MCP tools, skills, sandboxing, approvals, context management, and session state. The harness offers three interfaces: a chat UI, an HTTP API with a TypeScript SDK, and an embeddable UI SDK.

TrueForge offers flexibility in deployment, scaling from local mode, which runs on a single process with SQLite, to hosted mode with Postgres + Redis, Docker Compose, Helm, or Railway. The Quickstart guide assists in running TrueForge using these various methods.

To build and run an agent, one must connect models, tools, skills, and then create a reusable agent. TrueForge can be accessed and developed from the provided repository via the CONTRIBUTING.md file. Local mode is intended for personal use, offering an easy way to test TrueForge, but it lacks a login feature and stores data in a local SQLite file. Unauthorized data access or loss cannot be guaranteed, so it's advised to use local mode only on localhost.

For collaborative or production environments, hosted mode is recommended. TrueForge's performance has been benchmarked against Claude Managed Agents and deepagents, demonstrating comparable accuracy while reducing costs. Users interested in contributing may submit bug reports, feature requests, or documentation fixes through the CONTRIBUTING.md file and adhere to the Code of Conduct.

Forked pull requests should only modify the source code, and maintainers will regenerate the SDK upon merge. Security vulnerabilities should be reported via the Vulnerability Disclosure Program, as detailed in the SECURITY.md document.