tessera-llm-proxy released on PyPI for LLM cost optimization

tessera-llm-proxy hit PyPI on June 15 2026, letting developers add a single import to route OpenAI, Anthropic, Mistral, Groq, Cohere and other LLM clients through Tessera’s auto‑optimizing proxy [1].

The package installs with pip install tessera-llm-proxy and activates via import tessera; tessera.activate("tk_your_tessera_key"). Once active, the library patches supported SDK constructors at import time, so existing code—whether it calls openai.OpenAI(), anthropic.Anthropic(), or similar—continues to work unchanged while requests are automatically routed, cached, compressed and measured for cost savings. The proxy also supports frameworks that wrap these SDKs, such as LangChain or LlamaIndex, because they ultimately invoke the patched constructors.

Tessera offers a “Free Sandbox” tier that provides 60 million tokens per month, no credit‑card required, and a production tier that charges 20 % of measured savings, with a $100 minimum top‑up via Stripe. If the proxy delivers no savings, the fee drops to zero, and users can pause optimization at any time from the dashboard. The service supplies a dashboard key (tk_…) and optional environment‑variable configuration, and even lets developers tag requests for per‑feature savings reporting [1].

By exposing a lightweight, drop‑in proxy, Tessera aims to lower the barrier for cost‑conscious LLM developers who already have provider keys. The model shifts the risk of optimization to the provider—users only pay when savings materialize—while preserving existing authentication flows. Whether the proxy’s auto‑batching and compression deliver meaningful reductions at scale remains to be seen, but the free tier gives hobbyists a chance to test the claim without upfront cost.

The open question is how quickly developers will adopt the proxy and whether the measured savings justify the 20 % fee for production workloads, especially as LLM pricing evolves.