TypedAI is a TypeScript-first AI platform designed for developers, used to develop and run agents, LLM-based workflows, and chatbots. The platform includes highly capable software engineering agents, which have helped build the platform itself.
Main Capabilities
- Advanced autonomous agents: With reasoning/planning capabilities, inspired by papers such as Google's Self-Discover; supports memory and function call history, suitable for complex workflows; supports iterative planning and hierarchical task decomposition; can execute generated code in a sandbox for multi-step function calling and logic processing; LLM function schemas can be automatically generated from source code; supports human-in-the-loop for budget control, agent-initiated questions, and error handling.
- Software developer agents: Including a code editing agent (for local repositories), a software engineer agent (for ticket-to-merge-request workflows), a code review agent (configurable review guidelines, posting comments and suggested changes on the corresponding lines of GitLab merge requests), and an ad hoc repository query agent. The code editing agent can automatically detect project initialization, compilation, testing, and lint; the task file selection agent selects relevant files; the design agent creates an implementation plan; the code editing loop includes compilation, lint, testing, and fixes; the compilation error analyzer can search online, add additional files, and install packages; finally review the changes, and if necessary, perform additional code editing loops. The software engineer agent can find suitable repositories from GitLab/GitHub, clone and create a branch, invoke the code editing agent, and create a merge request.
- AI chat interface: Provides a chat interaction interface.
- Multi-LLM service support: Supports multiple LLM services such as OpenAI, Anthropic (native and Vertex), Gemini, Groq, Fireworks, Together.ai, DeepSeek, Ollama, Cerebras, SambaNova, OpenRouter, X.ai, and more.
- Multi-agent extended reasoning implementation: Provides a multi-agent extended reasoning implementation of the LLM interface.
- Configurable human-in-the-loop settings: Allows configuration of human participation steps.
- Function calling tools/integrations: Supports file system, Jira, Slack, Perplexity, Google Cloud, Gitlab, GitHub, etc.
- CLI and Web UI interfaces: Provides command-line and web interfaces.
- Deployment options: Can run locally, or be deployed to the cloud with multi-user/SSO.
- Observability: OpenTelemetry-based observability.
- Leverage the Python AI ecosystem: Leverage the broad Python AI ecosystem by executing Python scripts/packages.
Usage
TypedAI provides a powerful command-line tool for automation and development workflows:ai scripts run locally,aid run in Docker for isolation. Both can access all CLI agents, including the specialized codeAgent. For detailed CLI documentation, please refer to the CLI Usage Guide.
Run/deployment options include: running in single-user mode from a repository or a provided Dockerfile; achieving scale-to-zero deployment on Firestore and Cloud Run; multi-user SSO enterprise deployment (using Google Cloud IAP); Terraform, infrastructure scripts, and more authentication options coming soon.
To run a fully autonomous agent, use the corresponding command. LLM function calling patterns are automatically generated via the @func decorator on class methods, avoiding duplicate definitions using zod or JSON.
Applicable Scenarios
- Automated code editing tasks
- Software engineering workflows from tickets to merge requests
- Code review (configurable guidelines, posting comments on GitLab merge requests)
- Ad-hoc repository queries
- Building and running chatbots
- Developing LLM-based workflows
Limitations and Notes
TypedAI does not use LangChain; its platform scope covers the functionality found in LangChain and LangSmith. TypedAI code has the advantage of static typing, and prompt parameters can be statically typed, making refactoring easier; using simple control flow makes debugging via breakpoints/logs easier.
Regarding charges, pricing, specific platform support details, performance, etc., these are not clearly stated in the provided materials and are not provided for now.
