Hi everyone, today I’m super excited to announce Tessaract, a lightweight Python SDK that brings the most used frontier model APIs under a common interface. With Tessaract, switching between OpenAI and Anthropic models is as simple as changing 1 line of code. Tessaract unifies toolcalling, reasoning and multiturn agentic loops with a single interface across multiple model providers while retaining full support for caching and encrypted reasoning items enabling you to build your agent once and change your models anytime.
Every model has its strengths and weaknesses and the best agentic experiences are created by using different models for different tasks or building harnesses that allow users to switch models based on their tasks.
Building agents and applications that combine multiple frontier models is quite difficult because of the differences in request and response formats, tool schema definitions, reasoning configurations and other agentic AI primitives across OpenAI Responses API and Anthropic Messages API. This creates a lot of maintenance overhead, code bloat and repetitive logic rebuilding the same software infrastructure for each model provider. I faced these problems myself throughout my time building products at Brilliant AI.
I spent the last 4 months building Tessaract to solve these issues. Tessaract acts as a translation layer that sits on top of LLM APIs, taking your API calls and translating it to each API’s native format. This means that you get the same control, performance and behaviour you would expect from a native API call.
Tessaract is already being used in products like zain3d.com to power advanced pipelines for image to 3D scene generation.
You can find the documentation for tessaract on https://tessaractai.com which includes guides for making your first request and building an agent using tessaract. The project is fully open-source at https://lnkd.in/ePZNm7uC, please give the project a star if you like my work. I would love to hear your thoughts and feedback. Contributions are very welcome.
Hi everyone, today I’m super excited to announce Tessaract, a lightweight Python SDK that brings the most used frontier model APIs under a common interface. With Tessaract, switching between OpenAI and Anthropic models is as simple as changing 1 line of code. Tessaract unifies toolcalling, reasoning and multiturn agentic loops with a single interface across multiple model providers while retaining full support for caching and encrypted reasoning items enabling you to build your agent once and change your models anytime.
Every model has its strengths and weaknesses and the best agentic experiences are created by using different models for different tasks or building harnesses that allow users to switch models based on their tasks.
Building agents and applications that combine multiple frontier models is quite difficult because of the differences in request and response formats, tool schema definitions, reasoning configurations and other agentic AI primitives across OpenAI Responses API and Anthropic Messages API. This creates a lot of maintenance overhead, code bloat and repetitive logic rebuilding the same software infrastructure for each model provider. I faced these problems myself throughout my time building products at Brilliant AI.
I spent the last 4 months building Tessaract to solve these issues. Tessaract acts as a translation layer that sits on top of LLM APIs, taking your API calls and translating it to each API’s native format. This means that you get the same control, performance and behaviour you would expect from a native API call.
Tessaract is already being used in products like zain3d.com to power advanced pipelines for image to 3D scene generation.
You can find the documentation for tessaract on https://tessaractai.com which includes guides for making your first request and building an agent using tessaract. The project is fully open-source at https://lnkd.in/ePZNm7uC, please give the project a star if you like my work. I would love to hear your thoughts and feedback. Contributions are very welcome.