> ## Documentation Index > Fetch the complete documentation index at: https://aidocs.ethanbragdon.icu/llms.txt > Use this file to discover all available pages before exploring further. # Files & Context > Understand how Starfire AI uses conversation attachments, project context, and persistent knowledge sources. # Files & context Starfire can work with context at different lifetimes. Choosing the right layer keeps conversations more reliable and prevents important material from being trapped inside one chat. ## Three context layers ### Conversation context Use for temporary material that only matters to the current conversation. Examples: * an error screenshot * a configuration snippet * one document to summarize * a log file for a single debugging session ### Project context Use for material that belongs to an ongoing body of work. Examples: * product requirements * architecture notes * related conversations * build artifacts * project-specific instructions ### Knowledge context Use for reference material that should be indexed and retrieved repeatedly. Examples: * documentation collections * repositories * policies * research libraries * internal reference files ## Attachments are not automatically permanent knowledge Uploading a file to a conversation does not mean every future Starfire conversation should retrieve from it. Durable reuse belongs in a Project or Knowledge source. ## Context limits still matter Even when Starfire can access a large body of material, the active model has a finite context window. Retrieval systems help select relevant material rather than injecting every stored document into every request. ## File processing Depending on the file type and enabled features, Starfire can extract text, metadata, or other representations for use by a model or retrieval system. Processing status can matter: a newly uploaded file may exist before it is fully indexed for Knowledge queries. ## Sensitive files Only upload material you are authorized to process. Avoid putting passwords, API secrets, private keys, or unnecessary regulated data into ordinary AI context. A file attached to a chat, a file stored in a project, and a file indexed into Knowledge can have different retention, retrieval, and sharing behavior. Check the owning context before assuming who can access it. Learn how durable project context is organized. Learn how persistent retrieval and RAG sources work.