Home

Algenta Integrations

Framework integrations for Algenta — governed tool profiles, typed execution receipts, and policy-gated execution for the agent frameworks you already use. Ten packages, one shared contract, everything running against your own self-hosted engine. Apache-2.0.

Repository: thyn-ai/algenta-integrations · Contract: contracts/integration-tool-contract.json · Docs: docs.algenta.ai

Pick your path

"I want Algenta tools in my agent framework." Start with Getting Started (a complete LangChain walkthrough), then jump to your framework's guide: LangChain, LlamaIndex, Pydantic AI, Microsoft Agent Framework, Haystack, LiteLLM, Vercel AI SDK, n8n, vLLM, Ray Serve.

"I want to see governance proven, not promised." Run the conformance suite — twelve scenarios against a live engine, with an honest, falsifiable accounting of what passes (9), what is blocked (3), and why.

"I want to understand the governance model." Read the Tool Contract — the four profiles (observe / govern / execute / full), the never-model-facing fields, and the execution receipt.

"I want to deploy and operate this." Read Self-Hosted — endpoint resolution everywhere, why there is no hosted fallback, and the MCP vs HTTP surfaces.

"I want to contribute." Read Contributing — dev setup, the no-engine-dependency gate, and the release process. Questions? FAQ.

Packages

All ten packages are published — eight on PyPI, two on npm — and each carries its own passing test suite that runs against a real local stub server, never a mock of the framework's internals.

Package	Framework	Install	Guide
langchain-algenta	LangChain / LangGraph	pip install langchain-algenta	LangChain
llamaindex-algenta	LlamaIndex	pip install llamaindex-algenta	LlamaIndex
pydantic-ai-algenta	pydantic-ai	pip install pydantic-ai-algenta	PydanticAI
maf-algenta	Microsoft Agent Framework	pip install maf-algenta	MicrosoftAgentFramework
haystack-algenta	Haystack	pip install haystack-algenta	Haystack
litellm-algenta	LiteLLM (MCP Gateway)	pip install litellm-algenta	LiteLLM
ray-serve-algenta	Ray Serve / KubeRay	pip install ray-serve-algenta	RayServe
vllm-algenta	vLLM / OpenAI-compatible clients	pip install vllm-algenta	VLLM
algenta-tools	Vercel AI SDK (ai)	npm install algenta-tools	VercelAISDK
n8n-nodes-algenta	n8n	npm install n8n-nodes-algenta	N8N

What every package shares

• Four tool profiles decide which tools a model can even see — observe (read-only, the default), govern, execute, full. Same boundaries in every language.
• A typed execution receipt on a successful execute_decision; a blocked call is denied synchronously, by a named gate (idempotency, confidence, risk_floor).
• force / override_safety are never model-facing, in any profile — approvals are enforced by the engine server-side, never delegated to the model.
• Self-hosted only — every endpoint resolves from base_url= / ALGENTA_BASE_URL, with a localhost fallback. Nothing defaults to an Algenta-hosted cloud endpoint.

Status, stated plainly

• Every package is published and covered by its own real test suite.
• The conformance suite is 9 exercisable / 3 blocked today — see Conformance for the exact accounting. Until per-framework adapters drive the same scenarios through each package, no package here claims the "validated integration" label.
• A small number of engine capabilities (e.g. real timeout behavior) need a live deployment topology to verify. Where something is unverified, the docs say so.

Sibling repositories

• thyn-ai/algenta-sdk — the published Python and TypeScript client for the engine's public API.
• thyn-ai/mojo-kernels — clean-room Mojo kernels as drop-in accelerators for popular Python/TS libraries.

The Algenta engine itself is closed source and lives in a separate, private repository; these packages talk to it only through its public HTTP/MCP APIs.