OpenViking Memory Plugin for OpenCode

OpenCode plugin example that exposes OpenViking memories as explicit tools and automatically syncs conversation sessions into OpenViking.

Chinese install guide: INSTALL-ZH.md

Mechanism

This example uses OpenCode's tool mechanism to expose OpenViking capabilities as explicit agent-callable tools.

In practice, that means:

• the agent sees concrete tools and decides when to call them
• OpenViking data is fetched on demand through tool execution instead of being pre-injected into every prompt
• the plugin also keeps an OpenViking session in sync with the OpenCode conversation and triggers background memory extraction with memcommit

This example focuses on explicit memory access, filesystem-style browsing, and session-to-memory synchronization inside OpenCode.

What It Does

• Exposes four memory tools for OpenCode agents:
  • memsearch
  • memread
  • membrowse
  • memcommit
• Automatically maps each OpenCode session to an OpenViking session
• Streams user and assistant messages into OpenViking
• Uses background commit tasks to avoid repeated synchronous timeout failures
• Persists local runtime state for reconnect and recovery

Files

This example contains:

• openviking-memory.ts: the plugin implementation used by OpenCode
• openviking-config.example.json: template config
• .gitignore: ignores local runtime files after you copy the example into a workspace

Prerequisites

• OpenCode
• OpenViking HTTP Server
• A valid OpenViking API key if your server requires authentication

Start the server first if it is not already running:

openviking-server --config ~/.openviking/ov.conf

Install Into OpenCode

Recommended location from the OpenCode docs:

~/.config/opencode/plugins

Install with:

mkdir -p ~/.config/opencode/plugins
cp examples/opencode-memory-plugin/openviking-memory.ts ~/.config/opencode/plugins/openviking-memory.ts
cp examples/opencode-memory-plugin/openviking-config.example.json ~/.config/opencode/plugins/openviking-config.json
cp examples/opencode-memory-plugin/.gitignore ~/.config/opencode/plugins/.gitignore

Then edit ~/.config/opencode/plugins/openviking-config.json.

OpenCode auto-discovers first-level *.ts and *.js files under ~/.config/opencode/plugins, so no explicit plugin entry is required in ~/.config/opencode/opencode.json.

This plugin also works if you intentionally place it in a workspace-local plugin directory, because it stores config and runtime files next to the plugin file itself.

Recommended: provide the API key and tenant identity via environment variables instead of writing them into the config file:

export OPENVIKING_API_KEY="your-api-key-here"
export OPENVIKING_ACCOUNT="default"
export OPENVIKING_USER="opencode"

Configuration

Example config:

{
  "endpoint": "http://localhost:1933",
  "apiKey": "",
  "account": "default",
  "user": "opencode",
  "enabled": true,
  "timeoutMs": 30000,
  "autoCommit": {
    "enabled": true,
    "intervalMinutes": 10
  }
}

account and user are sent as X-OpenViking-Account and X-OpenViking-User tenant headers on every plugin API request; leave them empty to omit the headers.

The environment variables OPENVIKING_API_KEY, OPENVIKING_ACCOUNT, and OPENVIKING_USER take precedence over the config file.

Runtime Files

After installation, the plugin creates these local files next to the plugin file:

• openviking-config.json
• openviking-memory.log
• openviking-session-map.json

These are runtime artifacts and should not be committed.

Tools

memsearch

Unified search across memories, resources, and skills.

Parameters:

• query: search query
• target_uri?: narrow search to a URI prefix such as viking://user/memories/
• mode?: auto | fast | deep
• limit?: max results
• score_threshold?: optional minimum score

memread

Read content from a specific viking:// URI.

Parameters:

• uri: target URI
• level?: auto | abstract | overview | read

membrowse

Browse the OpenViking filesystem layout.

Parameters:

• uri: target URI
• view?: list | tree | stat
• recursive?: only for view: "list"
• simple?: only for view: "list"

memcommit

Trigger immediate memory extraction for the current session.

Parameters:

• session_id?: optional explicit OpenViking session ID

Returns background task progress or completion details, including task_id, per-category memories_extracted, and archived.

Usage Examples

Search and then read:

const results = await memsearch({
  query: "user coding preferences",
  target_uri: "viking://user/memories/",
  mode: "auto"
})

const content = await memread({
  uri: results[0].uri,
  level: "auto"
})

Browse first:

const tree = await membrowse({
  uri: "viking://resources/",
  view: "tree"
})

Force a mid-session commit:

const result = await memcommit({})

Memory Recall

The plugin can automatically search OpenViking memories and inject relevant context into each user message before it reaches the LLM. This uses OpenCode's chat.message hook to prepend a synthetic memory part to the outgoing message.

Note: This feature relies on OpenCode's chat.message hook contract. The hook signature or behavior may change in future OpenCode versions.

How It Works

1. On every user message, the plugin extracts text from the message parts
2. Searches OpenViking using semantic search (5-second timeout)
3. Ranks results using multi-factor scoring (base score + leaf boost + temporal boost + preference boost + lexical overlap)
4. Deduplicates results (abstract-based for regular memories, URI-based for events/cases)
5. Formats matching memories as a <relevant-memories> XML block
6. Prepends the block as a synthetic text part (synthetic: true) on the outgoing message so the memory persists across turns without polluting user input

If OpenViking is unavailable or the search times out, the message is passed through unchanged.

Recall Configuration

Add an autoRecall block to your openviking-config.json to customize recall behavior:

• enabled: boolean (default: true) — enable or disable automatic memory recall
• limit: number (default: 6) — maximum number of memories to inject (1–50)
• scoreThreshold: number (default: 0.15) — minimum relevance score for a memory to be included (0–1)
• maxContentChars: number (default: 500) — maximum characters per individual memory content
• preferAbstract: boolean (default: true) — prefer abstract (L0) content over full (L2) content when available
• tokenBudget: number (default: 2000) — approximate total token budget for injected memories (100–10000, estimated at 4 chars per token)

Example Config with Recall

{
  "endpoint": "http://localhost:1933",
  "apiKey": "",
  "account": "default",
  "user": "opencode",
  "enabled": true,
  "timeoutMs": 30000,
  "autoCommit": {
    "enabled": true,
    "intervalMinutes": 10
  },
  "autoRecall": {
    "enabled": true,
    "limit": 6,
    "scoreThreshold": 0.15,
    "maxContentChars": 500,
    "preferAbstract": true,
    "tokenBudget": 2000
  }
}

To disable recall, set "autoRecall": { "enabled": false }.

Notes for Reviewers

• The plugin is designed to run as a first-level *.ts file in the OpenCode plugins directory
• It intentionally keeps runtime config, logs, and session maps outside the repository example
• It uses OpenViking background commit tasks to avoid repeated timeout/retry loops during long memory extraction

Troubleshooting

• Plugin not loading: confirm the file exists at ~/.config/opencode/plugins/openviking-memory.ts
• Service unavailable: confirm openviking-server is running and reachable at the configured endpoint
• Authentication failed: check OPENVIKING_API_KEY or openviking-config.json
• No memories extracted: check that your OpenViking server has working vlm and embedding configuration