CopilotKit, the Seattle-based startup co-founded by Atai Barkai and Uli Barkai, released two infrastructure tools targeting production blockers in agentic AI development: AIMock for deterministic agent testing, and Pathfinder for self-hosted knowledge retrieval via MCP. The releases, detailed in a MarkTechPost analysis, extend CopilotKit’s 2026 shipping cycle alongside its AG-UI protocol, which handles the interaction layer between agents and human users inside applications.

AIMock: Fixture-Driven Testing for the Full Agent Stack

A single agent request in 2026 can touch six or seven services before returning a response: the LLM, an MCP tool server, a vector database, a reranker, a web search API, a moderation layer, and a sub-agent over A2A. Most teams mock one of them, according to MarkTechPost. The other six run live in CI, making test results non-deterministic.

AIMock replaces all of them from a single JSON configuration file and a single port. Coverage includes 11 LLM providers (OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure, Vertex AI, Ollama, and Cohere among them), full MCP JSON-RPC 2.0 mocking, A2A agent card discovery with SSE streaming, AG-UI event stream mocking for frontend tests, and vector database simulation compatible with Pinecone, Qdrant, and ChromaDB. The tool has zero external dependencies, built entirely from Node.js builtins.

Three capabilities separate AIMock from prior mocking tools. Record-and-replay proxies real API calls, saves them as fixtures, and replays them in CI indefinitely without touching live APIs. Drift detection runs daily against real provider APIs and catches response format changes within 24 hours, before developers encounter them in production. Chaos testing injects 500 errors, malformed JSON, and mid-stream disconnects to verify failure handling.

CopilotKit uses AIMock for its own AG-UI protocol end-to-end test suite, verifying agent behavior across LLM providers with fixture-driven responses.

Pathfinder is a self-hosted MCP server that indexes documentation, code, Notion pages, Slack threads, and Discord forums into searchable, agent-accessible knowledge. Configuration lives in a single pathfinder.yaml file. GitHub push events trigger incremental reindexing through webhook integration.

The search architecture combines hybrid vector and keyword retrieval. Pure semantic search fails on exact identifiers, error codes, and API names that appear verbatim in queries, a recurring problem for agent tools accessing technical documentation. Pluggable embeddings support OpenAI, Ollama, and local transformers.js, meaning fully air-gapped deployments require no external API key.

Auto-generated endpoints (/llms.txt, /llms-full.txt, and /.well-known/skills/default/skill.md) give agents standard discovery paths without additional configuration.

AG-UI’s Growing Ecosystem

The releases sit on top of CopilotKit’s AG-UI protocol, which defines the interaction layer between users, applications, and agents. AG-UI now has first-party SDK support from LangChain, CrewAI, Mastra, Agno, and Pydantic AI, with community implementations in Kotlin, Go, Dart, Java, Rust, Ruby, and C++, according to MarkTechPost. AWS integrated AG-UI into its FAST templates and Bedrock AgentCore. Google, Microsoft, Amazon, and Oracle all support the protocol.

The framing that once positioned MCP, A2A, and AG-UI as competing standards has given way to a layered model: MCP handles tool and data access, A2A handles agent-to-agent coordination, and AG-UI handles the presentation and interaction layer where users engage with agents.

The Testing Maturity Signal

The shift from “agent testing is theater” to fixture-driven, deterministic test infrastructure reflects a broader pattern: production-grade agent development requires the same CI/CD rigor as traditional software. AIMock’s approach, mocking every dependency an agent touches rather than just the LLM, makes the implicit argument that most teams shipping agents today are testing against live, non-deterministic services and calling it reliable.

CopilotKit’s GitHub repository hosts all three tools as open-source projects.