Supply the material needed for this task
Context includes your instructions, relevant evidence, conversation history and tool results. The model works with the material supplied to it during the task.
A knowledge store holds information between sessions. A context manager helps find and load the relevant parts. A folder full of notes becomes useful to the model when the agent retrieves the material it needs.
Keep evidence separate from instructions
State the task and its constraints. Then supply the source material the agent should use. Keep dates and source details with the evidence so that the agent can distinguish a current fact from an older record.
For a document review, that might mean the draft, the original notes and the audience brief. For a project handover, it might mean the goal, decisions already made and the next action.
Preserve decisions between sessions
Conversation history grows, and the amount a model can receive at once is limited. Save decisions and unfinished work in a form the next session can retrieve. Re-read the original evidence when a summary leaves a material question unresolved.
In Barnaby's AI Tools & Skills setup, notes live in Obsidian as local Markdown files. A hosted model receives the content loaded into its context. Local storage and local model inference are separate choices.
Context is one part of the harness. Connectors and tools explains how the agent retrieves material. On-device and cloud agents explains where processing takes place.