Building a coding agent from scratch
summary
Live-codes a working CLI coding agent that runs locally against Ollama with gpt-oss 20B, uses SSE HTTP client code, and raises the context window to 64K tokens so file reads fit. The agent separates reasoning from content (reasoning comes in a reasoning field or HTML tags), advertises tools to the model, executes tool calls and injects structured results back into history. Built tools include read/search/edit file tools (weather stubbed), and the demo works as a toy end‑to‑end.
AI-written summary · about
chapters (9)
- 0:00Building a coding agent
- 1:49Local model setup with Ollama
- 8:35Agent loop and input handling
- 12:14First LLM call and responses
- 20:19System prompts and reasoning
- 25:20Tool calls in the agent
- 31:20Dynamic tool registration
- 41:45File operations tools
- 48:10Summary and improvements
related talks
50mLet's build an AI agent
32mBuilding My Own Coding Agent
48mAI Coding Agents and How to Code Them
40mAI Coding Agents and how to code them
44mHow Does a Coding Agent Work?
48mAI Coding Agents and how to code them
01h 03mAI Coding Agents and how to code them
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips