Memory and Context Management

Intermediate⏱45 min📖3 min readLesson 5 of 6

Memory and Context Management

Manage memory and context in OpenCode sessions. Learn how to maintain state across conversations, use memory files, and optimize context for better results.

Understanding Context

Context is the information the AI remembers during a session:

Context Type	Description	Persistence
Conversation	Chat history	Session only
File Content	Read files	Until cleared
Project	Project structure	Session only
Memory Files	Persistent storage	Across sessions

Conversation Memory

View History

> /history

Save Session

> /save project-setup

Load Session

> /load project-setup

Clear Context

> /clear

Memory Files

Memory files persist information across sessions.

Project Memory

Create .opencode/memory/project.md:

markdown

# Project Memory

## Architecture
- Frontend: React with TypeScript
- Backend: Node.js with Express
- Database: PostgreSQL

## Conventions
- Use camelCase for variables
- Use PascalCase for components
- All API endpoints prefixed with /api/

## Decisions
- 2024-01-15: Chose PostgreSQL over MongoDB for ACID compliance

Personal Memory

Create ~/.config/opencode/memory/personal.md:

markdown

# Personal Memory

## Preferences
- Preferred model: gpt-4o
- Coding style: Functional programming
- Testing: Jest with 80% coverage minimum

## Shortcuts
- Use /clear between topics
- Always review AI suggestions before applying

Loading Memory

Automatic Loading

Configure auto-load in opencode.json:

json

{
  "memory": {
    "autoLoad": [
      ".opencode/memory/project.md",
      ".opencode/memory/conventions.md"
    ]
  }
}

Manual Loading

> Load the project memory file at .opencode/memory/project.md

Context Optimization

When Context Gets Too Large

Symptoms:

Slow responses
AI forgets early instructions
Token limit warnings

Solutions:

Clear and reload essentials:

> /clear
> Load .opencode/memory/project.md
> We're working on the authentication module

Use subagents for isolated tasks:

> Have the test-agent write tests for the login function

Summarize long conversations:

> Summarize what we've discussed so far into a memory file

Context Window Management

> /verbose
Context: 15234 tokens (47% of 32k limit)

Best Practices:

Situation	Action
Context > 70%	Clear and reload essentials
Switching topics	Clear conversation
Complex project	Use memory files
Multiple tasks	Use subagents

Memory File Structure

Recommended Format

markdown

# [Memory Type] Memory

## [Section]
- Key point
- Key point

## [Section]
- Key point

Naming Conventions

File	Purpose
`project.md`	Project architecture and decisions
`conventions.md`	Coding standards and patterns
`decisions.md`	Architectural decision records
`todo.md`	Current tasks and priorities
`learned.md`	Things learned during sessions

Using Memory in Prompts

Reference Memory

> According to our project memory, what database are we using?

Update Memory

> Update the project memory to note we've switched from MongoDB to PostgreSQL

Create from Session

> Save our conversation about the API design to .opencode/memory/api-design.md

Best Practices

Memory Hygiene

Practice	Benefit
Regular updates	Keeps memory accurate
Clear sections	Easy to find information
Date decisions	Track evolution
Remove outdated	Prevents confusion

Context Management

Practice	Benefit
Clear between topics	Prevents cross-contamination
Load essentials only	Reduces token usage
Use subagents	Isolates complex tasks
Monitor context size	Avoids limits