finds.dev← search

// the find

wrtnlabs/agentica

★ 1,027 · TypeScript · MIT · updated May 2026

TypeScript AI AI Function Calling Framework enhanced by compiler skills.

Agentica is a TypeScript framework that turns regular functions, OpenAPI specs, or MCP servers into LLM tool-calling agents without you writing any schema by hand. It uses the typia compiler to extract type information at build time, then handles the selector/executor/validator loop automatically. Aimed at TypeScript backend developers who want function-calling agents without the usual JSON schema boilerplate.

The typia-based compiler extraction is the real differentiator — you annotate TypeScript types normally and get LLM-ready schemas for free, which eliminates a whole class of schema drift bugs. The validation feedback loop (detect bad args, retry with correction) addresses the most common failure mode in function calling and is implemented as a first-class concern rather than an afterthought. The selector agent that filters candidate functions before passing them to the LLM is a practical solution to the context explosion problem when you have many tools registered. Multi-vendor JSON schema normalization (OpenAI uses a different subset of JSON Schema than Claude/Gemini) is handled transparently, which saves real pain when switching or multi-homing providers.

Heavy dependency on typia means you're locked into a specific compiler plugin setup — if your build toolchain doesn't support typia transformers, the core value proposition disappears and you're left with a generic function-calling wrapper. The vector-selector for function filtering introduces a pgvector or SQLite dependency for what is conceptually a routing concern, which is a significant operational footprint for smaller deployments. The monorepo is large and the package split (@agentica/core, rpc, vector-selector, benchmark, chat, cli) adds coordination overhead for contributors; the version bump story across packages isn't obvious. No clear story for stateful agents that need persistent memory across sessions — the history model appears to be in-process only.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →