Agentic AI / Agentic workflow
AI Coding Agent
An agentic coding assistant that plans tasks, inspects workspaces, calls tools, and coordinates local execution through a remote API.
- Streamlit
- FastAPI
- LangGraph
- MCP
- Render
- Multi-provider LLM fallback
Problem
Complex coding tasks need more than a conversational interface. An assistant must understand the workspace, choose tools, obtain approval for sensitive actions, and verify the result.
Approach
The system combines a Streamlit interface, FastAPI backend, task queue, LangGraph orchestration, MCP capabilities, and a local agent. Remote services coordinate the workflow while the local agent provides access to the user’s computer and workspace.
Architecture
Capabilities
- Natural-language coding tasks
- Task planning
- Workspace inspection
- File reading, creation, and editing
- Allow-listed command execution
- Tool calling and MCP
- Human-in-the-loop approval
- Verification and retry/error handling
- Task queue and progress tracking
- Chat threads and history
- Authentication
Overview
An agentic AI coding assistant built around a remote API and a local execution environment.
What it does
It handles natural-language coding tasks, plans work, inspects a workspace, reads and edits files, executes allow-listed commands, and tracks progress through a task queue.
Request lifecycle
A request enters through Streamlit, is coordinated by the FastAPI backend and task state, reaches the local agent, and is orchestrated through LangGraph and MCP before verification and response.
Planner and tool calling
The workflow separates planning from execution and uses tool calling to select workspace, file, command, and MCP capabilities.
LangGraph orchestration
LangGraph coordinates the agent path, execution steps, verification, retries, and error handling.
MCP architecture
The system supports MCP tools, resources, and prompts, including dynamic MCP tool discovery.
Safety and approval
Human-in-the-loop approval, command allow-lists, and workspace/path protection constrain actions that affect the local environment.
Local agent
The local agent is the boundary that provides access to the user’s computer and workspace. Complete coding execution depends on it.
Deployment
The Streamlit interface and FastAPI service are deployed remotely, with the API available through Render. The local agent is not replaced by that deployment.
Local setup
Local setup documentation and a verified local-run resource have not been supplied yet.