Everything is Context: Agentic File System Abstraction for Context Engineering
Context engineering—not model fine-tuning—should be the central challenge for generative AI systems, solved through a Unix-inspired file system abstraction that treats all context components uniformly.
Summary
This paper argues that managing external knowledge and information—context engineering—matters more than model fine-tuning for building effective AI systems. The authors propose treating context like files in Unix: a uniform abstraction that handles mounting, metadata, and access controls for diverse context sources.
The AIGNE framework implements this philosophy through a three-stage pipeline that emphasizes human oversight and accountability.
The AIGNE Pipeline
The framework processes context through three distinct stages:
flowchart LR
C[Constructors] --> L[Loaders]
L --> E[Evaluators]
C -.->|"Mount context sources"| L
L -.->|"Prepare context"| E
E -.->|"Assess quality"| Output[Agent Context]
- Constructors - Mount various context sources (documents, APIs, databases) using consistent interfaces
- Loaders - Transform and prepare context for consumption, handling metadata and access controls
- Evaluators - Assess context quality and relevance before injection into agent workflows
Core Insight
The Unix philosophy "everything is a file" provides a blueprint for context management. Just as Unix abstracts hardware devices, pipes, and files behind a uniform interface, AIGNE abstracts RAG pipelines, memory stores, tool outputs, and user inputs behind consistent mounting and access patterns.
This abstraction enables:
- Composability - Mix and match context sources without rewriting agent logic
- Traceability - Track which context influenced which decisions
- Access control - Apply permissions to context sources, not just model capabilities
Applications
The paper demonstrates two practical implementations:
- Memory-equipped agents - Persistent context that survives across sessions
- GitHub assistants - Repository context mounted and queried through the file system abstraction
Connections
- ai-engineering - Chip Huyen's book covers RAG and context management as core AI engineering skills; this paper proposes infrastructure-level abstractions for those techniques
- how-to-build-a-coding-agent - The agent loop described by Geoffrey Huntley allocates results back to context—AIGNE provides a formal model for that context management