TrueForge
TrueFoundry · Open-Source Runtime for Production AI Agents
TrueForge is an open-source agent harness for developers who need persistent sessions, MCP tools, sandboxed execution, approvals, and deployable APIs without committing to one model provider — positioned by TrueFoundry as a lower-cost, drop-in alternative to Claude Managed Agents.
Best for
Platform and application teams that want a model-neutral, self-hosted runtime for reusable production agents
Not ideal for
Nontechnical users seeking a ready-made assistant or teams unwilling to operate infrastructure and external model credentials
Who it's for
Developers and platform engineers building governed agents that must persist sessions, call tools, execute code, and deploy behind an API
Capabilities
- Runs model calls, MCP tools, skills, approvals, context management, and session state in one agent loop
- Supports OpenAI, Anthropic, Gemini, and OpenAI-compatible model endpoints
- Provides a bundled chat UI, HTTP API, TypeScript SDK, and embeddable UI SDK
- Uses persistent sessions and streamed events so applications can resume work and approval flows
- Runs code and file operations in isolated Daytona sandboxes with credentials held by the harness
- Deploys locally with SQLite or in hosted mode with Postgres and Redis (Docker Compose, Helm, or Railway)
- Schedules for unattended recurring runs (minimum interval one hour)
- Code Mode chains MCP tool calls in sandboxed Python; Generative UI and ask-user questions in chat
Limitations
- Local mode has no login by default and is explicitly not intended for production or internet-facing use
- Hosted deployments also have no login until you configure OIDC — anyone who can reach the URL is an admin
- Early-stage project (0.2.x releases as of September 2026), so APIs and packages are still changing
- Requires Node.js 22.14 or newer for the local npx setup
- Daytona is the only documented sandbox provider currently supported
- The free harness does not include model, sandbox, or infrastructure usage costs
Use cases
- Building a support agent that pauses before issuing refunds
- Deploying a research agent that delegates topics to parallel subagents
- Embedding a persistent agent session into an internal application
- Running the same agent across commercial, open, or OpenAI-compatible models
- Self-hosting a shared agent runtime with team authentication and durable storage
Our take
TrueForge is a runtime, not an orchestration framework: it covers what sits around the agent loop — sessions, approvals, sandboxes, APIs, and deployment — which libraries like LangGraph or CrewAI leave to you. That makes it most useful for platform teams that want to self-host one shared agent runtime and keep model choice open. It is early (0.2.x releases as of September 2026), sandboxing depends on Daytona, and hosted deployments have no login until you configure OIDC, so budget time for hardening. TrueFoundry's cost comparisons against Claude Managed Agents come from its own benchmark; measure on your own workload before relying on them.
Who should use it
Platform and backend teams building governed production agents that need persistent sessions, human approval gates, sandboxed code execution, and an HTTP API — without locking into one model provider.
Who should skip it
Non-developers looking for a ready-made assistant, teams that only need an orchestration library inside an existing app, and teams that cannot run Postgres and Redis or need a sandbox provider other than Daytona.
TrueForge pricing
Open Source
Free
- MIT-licensed harness
- Local and self-hosted deployment
- Bring your own model and infrastructure credentials
TrueFoundry Agent Harness (hosted)
Custom
- Managed multi-user hosting with auth and RBAC
- AI Gateway integration
- Standard pricing coming soon
Note: The harness is free; model, sandbox, and infrastructure charges are separate. TrueFoundry's hosted Agent Harness is custom-priced for now via sales.
Technical specs
Available models
TrueForge vs. competitors
TrueForge vs. LangGraph
LangGraph is a graph-oriented orchestration framework; TrueForge packages a broader runtime with sessions, UI, APIs, sandbox execution, and deployment modes.
TrueForge vs. CrewAI
CrewAI centers on role-based multi-agent workflows, while TrueForge centers on the production harness and infrastructure around reusable agents.
TrueForge vs. Dify
Dify emphasizes visual application building; TrueForge is a code-oriented runtime for teams that want to control models, tools, sandboxes, and hosting.
Frequently asked questions
What is TrueForge?
TrueForge is an open-source runtime that turns an LLM into a working agent by managing model calls, MCP tools, skills, sandboxing, approvals, context, and persistent session state.
Is TrueForge open source?
Yes. The TrueForge repository is released under the MIT License, so teams can inspect, modify, and self-host the harness.
Can TrueForge be self-hosted?
Yes. Local mode runs with SQLite for personal evaluation, while hosted mode uses Postgres and Redis through Docker Compose, Kubernetes (Helm chart), or Railway.
How much does TrueForge cost?
The open-source harness is free; you pay only for the models, sandbox provider, and infrastructure you connect. TrueFoundry's managed version, TrueFoundry Agent Harness, is custom-priced for now (contact sales), with standard pricing promised.
Integrations & fit
Alternatives to consider
LangChain
Developers building production multi-agent systems that need fine-grained control over state, execution flow, and human-in-the-loop checkpoints — and who are willing to trade setup time for that control
FreeCrewAI
Orchestrating autonomous agent teams for enterprise tasks
FreemiumLangGenius
No-code agentic workflows and RAG pipelines
FreemiumMastra
TypeScript-first AI agents and workflows for Node.js teams
Freemium
Cloudgeni
Product and platform teams building agent features into their own TypeScript/React product, or running agents for their organization, who want durable sessions, approvals, credential handling, and the choice to self-host under Apache-2.0 or start on a hosted cloud with the same API
Freemium
Docker
Developers and platform teams already using Docker who want versionable, shareable agents and multi-agent teams defined in config, with MCP tools and a free choice of cloud or local models
FreeAbout TrueForge
TrueForge provides the runtime layer between a model and a production agent. Teams configure models, MCP servers, git-backed skills, and a sandbox, then expose reusable agents through the bundled chat UI, HTTP API, TypeScript SDK, or embeddable UI. Its context controls include subagents, deferred tool loading, large-result offloading, and compaction, while approval gates pause sensitive tool calls for human review. Schedules run saved agents unattended on a recurring cadence (hourly at most), Code Mode lets the agent chain MCP calls in sandboxed Python, and the chat UI supports Generative UI. Local mode runs as one Node.js process with SQLite for personal evaluation; hosted mode uses Postgres and Redis through Docker Compose, Kubernetes (official Helm chart), or Railway. The harness is MIT-licensed and model-neutral, but teams still supply model credentials and currently need Daytona for sandbox execution. TrueFoundry's managed offering, TrueFoundry Agent Harness, embeds TrueForge unchanged and is currently custom-priced. TrueFoundry markets TrueForge as a drop-in alternative to Claude Managed Agents and reports, in its own benchmark, about 30% lower cost on the same model and about 75% lower with an open model.
Updates from TrueForge
TrueFoundry released TrueForge publicly with self-hosted and hosted deployment options, model portability, MCP tools, sandboxes, and approval workflows.
Are you the founder? Claim this listing →