Open-Mem for OpenCode

There is a particular kind of frustration in a long coding project: you explain the same decision to an assistant again because the useful context stayed in a previous session. open-mem is a community plugin for OpenCode that tries to address that gap by capturing activity, compressing it into observations, and recalling selected context later.

That convenience is also the reason to be careful. A memory plugin may see commands, file contents, and prompts that you would not normally store. I would test it on a small, non-sensitive project first and keep it only if the recalled context is more useful than the review and privacy burden.

Its capture, compression, storage, and configuration behavior can change between releases; verify the current plugin documentation before enabling it on an important project.

How the memory pipeline works¶

open-mem describes a three-phase workflow:

1. Capture — When you read files, run commands, or edit code, it captures outputs.
2. Compress — During idle time, it turns captures into structured observations.
3. Recall — In a later session, a compact index can be injected into the prompt.

Installation¶

Quick Start (Recommended)¶

npx open-mem

This adds open-mem to your OpenCode plugin config automatically. It starts capturing from your next session.

Manual Installation¶

bun add open-mem

Then add to your OpenCode config at ~/.config/opencode/opencode.json:

{
  "plugin": ["open-mem"]
}

AI Compression (Optional)¶

By default, open-mem uses a basic metadata extractor. For semantic compression, add an AI provider:

# Google Gemini (optional)
export GOOGLE_GENERATIVE_AI_API_KEY=your_key_here

Also supports Anthropic, AWS Bedrock, OpenAI, and OpenRouter. Auto-detects from environment variables.

Memory Tools¶

open-mem provides tools for interacting with project memories:

Finding Memories¶

Search past memories by query:

mem-find({ query: "authentication bug", limit: 5 })

Returns matching observations with IDs and summaries. Use mem-get to fetch full details.

Creating Memories¶

Save important observations:

mem-create({
  title: "Auth bypass vulnerability in login",
  type: "bugfix",
  narrative: "The login endpoint wasn't validating token expiration...",
  concepts: ["jwt", "authentication", "security"],
  importance: 10
})

Memory Types¶

Browsing History¶

View session timeline:

mem-history({ limit: 10 })

Drill into specific sessions using sessionId from the results.

Features¶

Hybrid Search¶

Combines FTS5 full-text search, vector embeddings (via sqlite-vec), knowledge graph traversal, and Reciprocal Rank Fusion. No external vector database needed.

Knowledge Graph¶

Automatic entity extraction with relationships. Graph-augmented search finds connections across sessions that keyword search would miss.

Progressive disclosure¶

A token-budgeted index can be injected into the system prompt. The agent sees what exists and decides what to fetch, reducing the need to load every observation at once.

Revision Lineage¶

Observations are immutable. Updates create new revisions that supersede the previous one. Deletes are tombstones with full audit trail.

Privacy and retention¶

The project documents local storage in .open-mem/ and redaction or exclusion options. Verify those guarantees in the installed release, add the directory to .gitignore, and inspect the database before sharing a repository. Local storage does not by itself prevent a configured AI provider from receiving data for compression.

mem-create({ narrative: "Used API_KEY_123 for testing" }) ```

Multi-Platform Support¶

open-mem isn't limited to OpenCode. Dedicated adapters bring the same capabilities to other tools:

Dashboard¶

Enable the web dashboard:

export OPEN_MEM_DASHBOARD=true
# Access at http://localhost:3737

The dashboard includes timeline, session, search, statistics, operations, and settings views. Treat its controls as release-dependent and verify the current documentation before relying on a particular workflow.

Configuration¶

Environment Variables¶

Custom Storage Location¶

export OPEN_MEM_PROJECT=/path/to/custom/project

Troubleshooting¶

Safe Database Reset¶

If you encounter database issues, use the maintenance CLI:

# Non-destructive WAL checkpoint
bunx open-mem-maintenance sqlite checkpoint --project /path/to/project --mode PASSIVE

# Non-destructive integrity check
bunx open-mem-maintenance sqlite integrity --project /path/to/project --max-errors 10

# Safe reset (blocked when active processes detected)
bunx open-mem-maintenance reset-db --project /path/to/project

Checking Health¶

For platform workers, check the queue mode:

{"command":"health"}

Returns status.queue.mode:

• enqueue-only: daemon healthy, worker signals PROCESS_NOW
• in-process: fallback mode when daemon unavailable

Comparison with Alternatives¶

Further Reading¶

• Getting Started
• Architecture
• Memory Tools Reference
• Search Documentation
• Configuration
• Privacy & Security
• Platform Adapters
• Troubleshooting

Official Links¶

• GitHub repository
• npm package
• Issue tracker