The agency bottleneck

The current state of AI agents is defined by a disconnect between reasoning and action. While large language models have become proficient at planning, the actual execution—connecting to a database, triggering a GitHub action, or querying a specific API—remains a fragmented mess of custom code and brittle integration logic. Every developer building an agentic application today is essentially reinventing the wheel for tool-calling infrastructure. Freight enters this space not by building agents themselves, but by building the pipes that connect those agents to the outside world.

A universal tool interface

Freight provides what it calls a universal tool layer. The core premise is simple: tools should be defined once and be accessible to any model or agent framework, whether it is LangChain, LlamaIndex, or a custom-built solution. In the current market, tool definitions are often tied to specific SDKs or model-specific JSON schemas. Freight abstracts this complexity, offering a hosted environment where tools live. This means a developer can deploy a tool, which might be a Python function or an API call, to Freight, and any agent can then invoke it through a standardized protocol. This approach mirrors how Stripe simplified payments; instead of managing the minutiae of bank gateways, developers interact with a clean abstraction.

Integration with Model Context Protocol

One of the most significant moves in the agent space recently is the introduction of the Model Context Protocol (MCP) by Anthropic. Freight is a core player in this emerging standard. By acting as a host for MCP servers, Freight allows developers to bridge the gap between local tools and cloud-hosted agents. This is a critical piece of the puzzle for enterprise adoption. It allows agents to interact with secure, internal data through a standardized interface without requiring the agent itself to have direct, unmediated access to every internal system. Freight's infrastructure handles the translation between what the model wants to do and what the protocol requires.

Managing the mechanics of action

Beyond simple connectivity, Freight addresses the thorny issue of authentication and permissions. When an agent needs to perform an action on behalf of a user, such as sending an email or updating a CRM, it needs access to that user's credentials. Managing these OAuth flows and session tokens within an agentic loop is notoriously difficult and poses a security risk. Freight architecture treats authentication as a core feature, managing the secure storage and injection of credentials into tool calls. This allows the agent logic to remain focused on the task at hand rather than the mechanics of API headers.

Market positioning and the developer stack

Founded in 2024 and based in San Francisco, Freight is led by a team with deep infrastructure roots. Yining Zhao and Sushrut Upadhyaya, both with backgrounds in scaled engineering environments, are building for a world where agentic capabilities are standard features of every application. They are competing in a nascent category alongside players like LangChain or various agent-cloud startups, but Freight focus is narrower and more infrastructure-centric. They are not trying to be the brain; they are the hands and the nervous system. For developers, this represents a shift from building monolithic agent scripts to assembling modular systems where the capabilities are decoupled from the underlying model intelligence.