TrueFoundry, a startup founded by former Meta engineers, has launched TrueForge, an open-source AI agent harness designed to give developers more control over AI agents and tools while significantly reducing costs. Released under the permissive MIT License on GitHub, TrueForge allows for customisation, forking, self-hosting, and integration into commercial products, offering a vendor-neutral approach to AI agent management.
The company claims impressive cost savings, stating that using TrueForge with the open-source GLM-5.2 LLM to complete tasks on DevRev’s Enterprise-Bench was 75% cheaper than using Anthropic's Claude Managed Agents. Even when using the same model, Anthropic's Opus 4.8, TrueForge reportedly offers a 30% cost reduction compared to Claude Managed Agents. This efficiency is attributed to TrueForge's focus on context management, including delaying tool schema loading, delegating tasks to subagents, offloading large results to files, processing structured outputs with code, and compacting conversations. It also features a unique sandbox-as-a-tool architecture that reduces unnecessary compute by only provisioning a sandbox when code execution or file manipulation is required.
TrueForge is positioned to complement TrueFoundry's existing commercial AI Gateway, which handles centralised control of model access, credentials, and budgets. While the open-source harness can be used independently, its true enterprise value is realised when integrated with the AI Gateway, which provides robust governance, single sign-on (SSO), and granular permission enforcement. This approach allows enterprises to manage AI agents and tools under a unified control plane, regardless of the specific harnesses or models employed. Beta users like NetApp have leveraged TrueForge for incident response and ticket triaging, as well as exposing internal agents as self-service tools.
Fuente Original: https://venturebeat.com/orchestration/truefoundrys-open-source-ai-agent-harness-trueforge-boasts-30-75-cheaper-task-completion-than-claude-managed-agents
Artículos relacionados de LaRebelión:
- Open Weights Are Driving AI Development Forward
- Local AI Powerhouse Qwen38-27B Runs Frontier Agents Locally
- AI Agents Fight Conflicting Instructions Lead to Sabotage
- X Unveils Algorithm Secrets Ranking Filtering Now Open
- Capital Ones AI Strategy Open Models Custom Data Power
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario