Tesserix Agent Development Kit documentation

Tesserix Agent Development Kit is the reusable Python runtime and integration layer for agents that need typed contracts, provider portability, production controls, and network-free tests. The application owns its process and deployment; the kit supplies composable primitives.

Choose a path

Goal	Start here
Run the offline example	Getting started
Browse every implemented capability	Capability map
Use the installed project commands	Command-line guide
Copy an executable example	Runnable cookbook
Build a new custom agent	Build a custom agent
Design the full path from source to monitored production	Agent lifecycle and platform architecture
Connect OpenAI, Anthropic, Gemini, Groq, Grok, OpenRouter, or a local model	Provider recipes
Connect a gateway, MCP server, store, or workflow engine	Integrations and gateways
Reuse a tool or agent from another framework	Framework interoperability
Publish or consume an official A2A Agent Card	Official A2A interoperability
Connect a Tesserix agent to Google Agent Development Kit	Google Agent Development Kit bridge
Test an agent without spending money	Testing
Keep a deployed agent on a current reviewed kit release	Keep agents current safely
Understand contribution and branch protections	Repository governance
Decide whether the project is ready for a public or production rollout	Public-readiness review
Read the community standards or report conduct privately	Code of Conduct
Contribute a change	Contributing

Mental model

An application composes four independent pieces:

1. An Agent declares the job, model or task class, tool allowlist, answer shape, and policy.
2. A ModelProvider translates one model API into the kit's request, response, usage, stream, capability, and error types.
3. A ToolRegistry holds typed functions and produces a fixed per-agent view.
4. AgentRunner applies boundaries and drives the model/tool loop to a terminal Run.

That separation is the portability mechanism. A provider, model gateway, store, MCP server, or A2A registry can be replaced at its boundary without giving the agent vendor objects.

For the production sequence—evaluation gates, approved registry versions, canary, execution planes, sandboxes, recovery and feedback—read Agent lifecycle and platform architecture.

Main references

• Capability map
• Command-line guide
• Runnable cookbook
• Architecture
• Agent lifecycle and platform architecture
• Core primitives
• Providers and capabilities
• Tools
• Run loop
• Tenancy
• Security and threat model
• Budgets
• Guardrails
• Memory
• Retrieval
• MCP client
• AgentGateway
• Google Agent Development Kit bridge
• Framework interoperability
• Tesserix peer discovery
• Durable runs
• Evaluations
• Observability
• Stability
• Release verification
• Keep agents current safely
• Repository governance

What the name means

Google also has an Agent Development Kit. This project's distribution is tesserix-adk, its import namespace is tesserix_adk, and its interoperability adapter targets the independent Agent2Agent protocol through the official a2a-sdk. The two projects can work together through that boundary.

The project is pre-1.0. Consult Stability before adopting an alpha subpackage in a long-lived integration.