tessera-autogen 0.1.0

Drop-in Tessera integration for AutoGen (autogen_core / autogen_agentchat / autogen_ext). One line of config routes your AutoGen agents' LLM calls through Tessera's auto-route + auto-cache + auto-compress + auto-batch proxy. Free 60M tokens/mo. Production: 20% of measured savings.

tessera-autogen

Drop-in Tessera integration for AutoGen 0.4+ (autogen_core / autogen_agentchat / autogen_ext). One line of config routes every LLM call your AutoGen agents make through Tessera's auto-route + auto-cache + auto-compress + auto-batch proxy.

Free 60M tokens/month. Production: 20% of measured savings. No card up front.

Install

pip install tessera-autogen

Requires Python 3.10+. autogen-core + autogen-ext are peer dependencies — install them alongside this package.

Usage

The most common pattern uses a bundled factory to construct a pre-wired AutoGen ChatCompletionClient:

from autogen_agentchat.agents import AssistantAgent
from tessera_autogen import tessera_openai_client

client = tessera_openai_client(
    model="gpt-4o",
    openai_api_key="sk-...",   # your OpenAI key
    tessera_api_key="tsr_...",  # get a free one at tesseraai.io/dev
)

agent = AssistantAgent(name="researcher", model_client=client)

# Rest of your AutoGen code runs unchanged — single-agent calls,
# SelectorGroupChat, Swarm, tool use all route through Tessera and
# benefit from auto-optimization.

For Anthropic models:

from tessera_autogen import tessera_anthropic_client

client = tessera_anthropic_client(
    model="claude-sonnet-4-6",
    anthropic_api_key="sk-ant-...",
    tessera_api_key="tsr_...",
)

For explicit ChatCompletionClient construction (rare — useful when you need fine-grained client kwargs):

from autogen_ext.models.openai import OpenAIChatCompletionClient
from tessera_autogen import tessera_openai_config

client = OpenAIChatCompletionClient(
    model="gpt-4o",
    api_key="sk-...",
    **tessera_openai_config(api_key="tsr_..."),
)

What Tessera does for your AutoGen workloads

• Auto-route — calls to expensive models are evaluated for a cheaper alternative that preserves quality on canary samples.
• Auto-cache — exact-match + semantic cache for repeat queries. Multi-agent loops often re-issue identical sub-prompts; cache returns are free.
• Auto-compress — system prompts + tool descriptions compressed losslessly before they hit the upstream provider.
• Auto-batch — async teams with batch-tolerant SLAs get arbitraged onto provider batch APIs for ~50% cost reduction.

All gated by per-workload quality canaries; toggle any mechanic on/off from the Tessera dashboard. Free Sandbox tier gives you observe-only mechanics; Production tier unlocks the full stack.

Supported providers (v0.1)

Provider	Status	Config function
OpenAI	✅ verified	tessera_openai_config, tessera_openai_client
Anthropic	✅ verified	tessera_anthropic_config, tessera_anthropic_client
Mistral / Gemini	🚧 queued for v0.2	—

v0.1 covers ~85% of customer traffic per our outreach research. Open an issue if you need a provider on the queue surfaced sooner.

Companion packages

Companion to tessera-sdk (vanilla provider SDKs), tessera-langchain (LangChain integration), tessera-vercel-ai (Vercel AI SDK integration), tessera-llamaindex (LlamaIndex integration), tessera-mastra (Mastra Agent framework integration), tessera-pydantic-ai (Pydantic AI integration), and tessera-crewai (CrewAI multi-agent integration). Same proxy, same mechanic stack, AutoGen-shaped API.

License

Apache 2.0 — see LICENSE.