Building AI Agents Using Virtual File Systems and Bash
cjbell · hn · 2026-07-09
Core Philosophy: Files Over Tools
The Knock team shares their engineering experience in building AI agents. They abandoned the approach of having LLMs directly call various complex external API tools, switching to an architecture combining a Virtual File System (VFS) with Bash.
Architectural Advantages and Implementation
- Reduced Complexity: Unifying various operations into VFS reads/writes and Bash command executions drastically reduces the number of tools the model needs to adapt to, along with complex JSON Schema definitions.
- Enhanced Reliability: Models generally possess a deeper and more stable understanding of file systems and command-line operations than specific API calls, thereby lowering tool-call hallucinations and failure rates.
- State Management: Through a VFS, agents can easily pass, save, and read intermediate state data across different steps, much like humans use local files.
This "files-first" paradigm offers an inspiring new approach for building complex, highly reliable enterprise-level Agents.
More from coding & agent
- HeyGen adds a media-sourcing skill for coding agents with 75k images and 10k tracks — HeyGen · 2026-07-22
- Agent search bottlenecks are now about variance, not raw latency — rohanpaul_ai · 2026-07-22
- LangSmith adds tracing for Pipecat, LiveKit, OpenAI Realtime, and Gemini Live — LangChain · 2026-07-22
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22