Agentica — AI Function Calling Framework
Turn TypeScript classes, Swagger/OpenAPI documents, or MCP servers into powerful AI agents without writing JSON Schema by hand.
- Source repo
- wrtnlabs/agentica
- Stars
- ★ 1k
- Last updated
- 5mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 49/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesChatGPT · Claude.ai · OpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Developers who want to expose existing TypeScript classes as AI-agent functions without manually writing JSON Schema.
- Not a fit if
- Plain JavaScript projects unwilling to add a typia compile step
- Teams needing built-in cost control or budget management
- Source review
- 49/100 · Major gaps 2 safety controls not found
What does this agent do, and when should you use it?
Agentica is a TypeScript framework specialized in AI Function Calling, developed and open-sourced by Wrtn Technologies. It enables developers to build AI agents from three sources: TypeScript classes, Swagger/OpenAPI documents, and MCP servers, automatically generating function-calling schemas via compiler skills (e.g., typia). The framework handles specification differences across LLM vendors (OpenAI, Google Gemini, Anthropic Claude, DeepSeek, Meta Llama) automatically. It provides a CLI setup wizard (npx agentica start) to scaffold projects, and includes a WebSocket protocol for client-server communication. Key features include validation feedback to correct AI errors, and a selector agent to reduce token consumption and improve accuracy.
Agentica reads TypeScript classes, Swagger/OpenAPI documents, or MCP server definitions and uses typia's compiler to generate function-calling schemas automatically. Developers instantiate an Agentica object, configure the vendor (e.g., OpenAI) and controllers (e.g., typia.llm.controller for TypeScript classes, assertHttpController for OpenAPI). Calling agent.conversate() processes user input, selects candidate functions, executes calls, and uses validation feedback to fix mistakes. It supports multiple vendors and custom JSON Schema. The CLI scaffold creates projects with WebSocket support for deployment.
- Developers who want to expose existing TypeScript classes as AI-agent functions without manually writing JSON Schema.
- Backend teams leveraging existing Swagger/OpenAPI docs to build chat agents or automated services.
- Teams needing a unified interface across multiple LLM providers (e.g., OpenAI and Anthropic) for function calling.
- Building enterprise e-commerce agents that search products, place orders, or manage inventory through function calls.
- Integrating MCP servers to extend AI agents with external tools and data sources.
- Organizations seeking a more flexible alternative to rigid workflow-based agents.
How do you install or deploy this agent?
Run npx agentica start <directory> to launch the setup wizard. Choose a package manager (npm, pnpm, yarn), project type (NodeJS, NestJS, React, or Standalone), and embedded controllers. The wizard creates a project skeleton. For existing projects, install @agentica/core via npm and ensure typia and OpenAI SDK are also installed.
How do you use this agent?
Import { Agentica, assertHttpController } from '@agentica/core'. Create an Agentica instance with vendor configuration (e.g., OpenAI API key and model) and controllers. For TypeScript classes, use typia.llm.controller; for OpenAPI documents, use assertHttpController with a fetched Swagger JSON. Call await agent.conversate('your query') to interact. For deployment, non-Standalone project types enable WebSocket protocol for client-server communication.
What are this agent's strengths and limitations?
- Compiler-driven schema generation eliminates manual JSON Schema authoring errors and speeds up development.
- Automatic adaptation to different LLM vendor specifications (OpenAI, Google, Anthropic, DeepSeek, Llama) reduces integration overhead.
- Validation feedback mechanism detects and corrects AI mistakes in argument composition, improving reliability.
- Selector agent filters candidate functions to minimize context usage and token consumption, optimizing performance.
- Supports three function sources: TypeScript classes, OpenAPI documents, and MCP servers, offering flexibility.
- Primarily targets TypeScript developers; requires familiarity with TypeScript and typia.
- Dependence on typia compiler means an extra build step and potential learning curve.
- Setup wizard may generate unnecessary boilerplate for simple projects, adding complexity.
- No built-in cost control; users must manage API keys and vendor pricing themselves.
- Focuses on conversational interactions, so complex long-running tasks may need custom orchestration.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Agentica — AI Function Calling Framework This agent | 49 · Major gaps | CLIFree + model costs | ★ 1k | 5mo ago | TypeScript | ChatGPT · Claude.ai · OpenAI API · Claude API |
| Corsair Integration Platform | 53 · Major gaps | Library / SDKFreemium | ★ 15k | today | TypeScript | — |
| OpenConnector | 57 · Major gaps | Self-hosted serviceFreemium | ★ 6k | today | TypeScript | — |
| Composio | 51 · Major gaps | Library / SDKFree + model costs | ★ 31k | 2d ago | TypeScript | Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README demonstrates calling external APIs via HTTP controllers, but no mention of least privilege or user confirmation. Test code uses API keys but does not explain secure handling. Dependencies are locked in pnpm-lock.yaml, but no security audit is provided. External effects (e.g., HTTP calls) exist, but no rollback mechanism is provided. Source attribution: README and package.json clearly attribute to Wrtn Technologies, but publisher is unverified. Deductions: lack of user confirmation, rollback, and insufficient sensitive data handling details.
Evidence shows: README and package.json consistently describe an AI function calling framework. Dependencies are managed via pnpm with lockfile, but no dependency availability guarantee. Failure messages: test code has error handling, but no user-facing failure messages. Deduction: insufficient failure messages.
Evidence shows: README provides multiple usage scenarios (TypeScript classes, Swagger, MCP) and offers playground and tutorials. Capability boundaries: clearly states support for three protocols. Trigger precision: defined via function lists, but no precise trigger conditions. Environment fit: supports Node.js and multiple package managers, but no browser support mentioned. Deductions: trigger precision and environment fit details are thin.
Evidence shows: README provides installation, usage, and documentation links, with clear information architecture. Install notes: provides npx agentica start command. Naming stability: package names like @agentica/core, but no naming change history. Examples and FAQ: multiple examples, but no FAQ. Known limitations: not explicitly listed. License: MIT. Versioning and changelog: package.json has version, but no changelog. Maintenance responsibility: README states maintained by Wrtn Technologies. Deductions: missing known limitations and changelog.
Evidence shows: README claims improved ease and stability of function calling, with comparison table. Output usability: provides API docs and examples. Marginal value: improvements over vanilla function calling. Cost-benefit: claims reduced token consumption, but no specific data. Deduction: cost-benefit lacks data support.
Evidence shows: README claims (e.g., improved stability) lack specific evidence. Cross-source corroboration: has YouTube tutorials and Discord community, but no independent verification. Fact-inference separation: README distinguishes facts and inferences, but not clearly. Deductions: claims lack traceability.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Publisher identity is unverified; proceed with caution.
- No user confirmation mechanism; external actions may be executed automatically.
- Sensitive data handling details are insufficient; API key management is user's responsibility.
- No rollback mechanism; failed external calls may not be recoverable.
- Dependency security audit is missing; check dependencies for vulnerabilities yourself.