WorkGPT

WorkGPT enables LLMs to invoke arbitrary APIs by converting OpenAPI/JSON schemas into function definitions that the model can call. The framework parses API specifications, generates function signatures, and routes LLM-selected function calls to actual HTTP endpoints with parameter binding and response handling. This allows agents to dynamically discover and invoke external services without hardcoded integrations.

WorkGPT implements an agentic loop that iteratively prompts the LLM to select from available tools/APIs, executes the chosen action, and feeds results back into the model for next-step planning. The framework manages conversation state, tracks tool invocation history, and implements stop conditions (max iterations, goal completion). This enables complex workflows where the model autonomously chains multiple API calls to accomplish user objectives.

WorkGPT automatically parses API responses (JSON, XML, plain text) and injects them back into the LLM context for further reasoning. The framework handles response formatting, truncation for large payloads, and type conversion to ensure the model receives usable data. This enables the agent to reason about API results and decide on subsequent actions based on actual response content.

WorkGPT provides a templating system for constructing agent prompts that include available tools, instructions, and context. The framework manages system prompts, tool descriptions, and user input formatting to ensure the LLM receives well-structured instructions for tool selection and reasoning. This enables consistent agent behavior and makes it easy to modify instructions without changing core agent logic.

WorkGPT abstracts away provider-specific API differences through a unified interface, allowing agents to switch between OpenAI, Anthropic, and other LLM providers without code changes. The framework handles provider-specific function calling formats, parameter mapping, and response parsing. This enables portability and cost optimization by allowing runtime model selection.

WorkGPT implements error handling for API failures, timeouts, and malformed responses, with configurable retry strategies and fallback behaviors. The framework catches HTTP errors, network timeouts, and parsing failures, then either retries the request or informs the agent of the failure for alternative action selection. This improves agent robustness when dealing with unreliable or slow APIs.

WorkGPT supports multiple authentication methods (API keys, OAuth2, basic auth, custom headers) and manages credentials securely without exposing them in prompts or logs. The framework handles credential injection into API requests and supports environment variable-based configuration for secure credential storage. This enables agents to authenticate with protected APIs while maintaining security.

WorkGPT logs all agent actions, API calls, and LLM responses for debugging and monitoring. The framework captures tool selection reasoning, API request/response pairs, and execution timing, making it easy to understand agent behavior and diagnose failures. Logs can be exported for analysis or integrated with external monitoring systems.

WorkGPT can dynamically load API schemas from OpenAPI endpoints, files, or registries, allowing agents to discover available tools at runtime without hardcoding tool definitions. The framework parses schemas, extracts function signatures, and updates available tools without restarting the agent. This enables flexible agent configurations and integration with API catalogs.

WorkGPT validates function parameters against schema definitions and coerces types (string to integer, array formatting, etc.) to match API expectations. The framework checks required parameters, validates against constraints (min/max, enum values), and provides clear error messages when validation fails. This prevents invalid API calls and improves agent reliability.