install-openviking-memory

Imported from volcengine/openviking (examples/openclaw-plugin/skills/install-openviking-memory/SKILL.md). Install upstream with npx skills add volcengine/openviking --skill install-openviking-memory. Copyright stays with the author.

Install OpenViking Memory — Agent Installation & Reference Skill

You are about to help a user install and configure the OpenViking memory plugin for OpenClaw. The user does NOT know how to use a terminal. You will collect a few values through natural conversation, then run all commands yourself using your shell tool.

Read this entire document before starting. Follow every step in order. Do NOT skip steps.

This skill is split into two halves:

• Part 1 (STEP 0–10): Natural-language installation walkthrough. Use this on first install or reconfigure.
• Part 2: Reference. Tools, config schema, multi-tenant, multi-instance, daily ops, uninstall, error recovery. Read on demand.

Server scope. This skill does not install the OpenViking server itself. It assumes the server is already running locally on 127.0.0.1:1933 or on another machine. If the user has no server, see "Server not ready" handling in STEP 5.

Part 1 — Natural-Language Installation Walkthrough

STEP 0: Detect Language

If the user's first message contains Chinese characters, respond in Chinese throughout. Otherwise respond in English. All user-facing messages below have (CN)/(EN) variants — use the matching one.

Do NOT show this step to the user.

STEP 1: Detect Operating System (silently)

Run this command and remember the result. Do NOT show it to the user.

uname -s 2>/dev/null || echo WINDOWS

• Output contains Darwin → OS=mac
• Output contains Linux → OS=linux
• Output is WINDOWS or the command fails → OS=windows

STEP 2: Check Prerequisites (silently)

Run silently:

node -v
openclaw --version

If node is missing:

(CN) 你的系统没有安装 Node.js。OpenClaw 和 OpenViking 插件需要 Node.js >= 22。请先安装 Node.js，然后再回来找我。 (EN) Node.js is not installed. OpenClaw and the OpenViking plugin require Node.js >= 22. Please install Node.js first, then come back.

Stop. Do NOT continue.

If openclaw is missing:

(CN) 你的系统没有安装 OpenClaw。请先安装 OpenClaw（>= 2026.5.27），然后再回来找我。 (EN) OpenClaw is not installed. Please install OpenClaw (>= 2026.5.27) first, then come back.

Stop. Do NOT continue.

If both exist, proceed to STEP 3 silently.

STEP 3: Greet and Ask for 3 Values

Send this message:

(CN) 好，我来帮你接入 OpenViking 长期记忆。装好之后，我会自动记住对话里的重要信息，下次聊也能回忆起来。

我需要 3 条信息，不知道的可以问你的管理员：

1. OpenViking 服务地址 —— 例如 https://ov.example.com 或 http://192.168.1.100:1933，本机服务可以直接说"本机"
2. API Key —— 用来鉴权；服务没开认证可以说"没有"
3. 记忆归属（可选） —— 所有对话共用 user 记忆（默认），还是在 peers/<assistant_id> 或 peers/<sender_id> 下分开存放

先告诉我服务地址吧？

(EN) I'll set up OpenViking long-term memory for you. Once configured, I'll automatically remember important info from our chats and recall it later.

I need 3 things (ask your admin if unsure):

1. OpenViking server URL — e.g. https://ov.example.com or http://192.168.1.100:1933. For a local server, just say "local".
2. API Key — for auth. Say "none" if the server has no auth.
3. Memory scope (optional) — shared user memory (default), or peer memory under peers/<assistant_id> / peers/<sender_id>.

What's the server URL?

STEP 4: Collect Values

Collect 3 values through natural conversation. Be flexible: if the user gives several at once, parse them all. If they correct something, accept the new value.

4a. BASE_URL (REQUIRED)

• "local" / "本机" / "localhost" → use http://127.0.0.1:1933.
• ov.example.com without protocol → prepend https://.
• Strip trailing /, /health, or /api.
• After normalization must start with http:// or https://.
• If the user says they don't know, ask them to check with the admin or look at how the server was started. Do NOT make up a URL.

4b. API_KEY (OPTIONAL)

(CN) API Key 是什么？服务没开认证就直接说"没有"。 (EN) What's the API Key? Say "none" if the server has no auth.

• "no" / "none" / "没有" / "空" / empty → API_KEY="" (you will skip the flag later).
• Otherwise store as-is.

4c. PEER_ROLE (OPTIONAL)

(CN) 记忆要存在哪一层？默认共享在 viking://user/<user_id>/memories（none）。也可以按 OpenClaw 助手存到 .../peers/<assistant_id>/memories（assistant），或按发送者存到 .../peers/<sender_id>/memories（sender）。 (EN) Where should memory live? The default is shared viking://user/<user_id>/memories (none). You can instead use .../peers/<assistant_id>/memories (assistant) or .../peers/<sender_id>/memories (sender).

Default to none. Never silently pick another value — the three options mean different storage layouts and are awkward to change later:

person is a legacy alias for sender. Accept it from an existing config or explicit user input, but normalize/write new setup values as sender.

OpenViking initializes the managed peers/ container for every user. Do not tell the user that none removes that container: it only avoids a concrete peer memory subtree. With assistant or sender, actor-peer recall includes shared user memory plus the current peer memory; it hides other peer subtrees. Changing this setting does not move existing memories.

• Empty / "default" / "默认" / "共用" → leave unset (plugin defaults to none).
• Otherwise pass --peer-role none|assistant|sender. Treat explicit legacy person as sender. Any other value → ask again.

4d. PEER_PREFIX (OPTIONAL, only when PEER_ROLE=assistant)

(CN) 想给这个 agent 一个记忆前缀吗？留空就用默认。只能用字母、数字、_、-。 (EN) Want to set an peer prefix? Leave blank for the default. Letters, digits, _, - only.

• Only meaningful when PEER_ROLE=assistant; skip the question otherwise.
• Empty / "default" / "默认" → leave unset (plugin defaults to "").
• Otherwise validate against /^[A-Za-z0-9_-]+$/. If invalid, ask again.

4e. (Conditional) Multi-Tenant Root-Key Fields

Only ask for these if STEP 7 detects a root key (Root API key detected. Missing: --account-id, --user-id). Don't ask up front.

• ACCOUNT_ID
• USER_ID

See Reference: Multi-Tenant for what these mean.

STEP 5: Pre-flight Connectivity Check (silently)

Tell the user briefly:

(CN) 我先测一下能不能连上服务…… (EN) Let me test the connection to your server...

Run:

If OS=windows:

try { (Invoke-WebRequest -Uri "BASE_URL/health" -TimeoutSec 10 -UseBasicParsing -ErrorAction Stop).StatusCode } catch { $_.Exception.Response.StatusCode.value__ }

If OS=mac or OS=linux:

curl -sS -o /dev/null -w "%{http_code}" --connect-timeout 10 "BASE_URL/health"

Replace BASE_URL with the actual value.

Server-not-ready handling

This skill does not install or operate the OpenViking server. If the user's server is unreachable, present the situation honestly and offer two paths:

(CN) ❌ 我连不上 BASE_URL。可能是：

1. 服务还没启动 —— 请联系你的 OpenViking 服务管理员把它起起来；如果是你自己负责，请参考 OpenViking 官方文档（https://github.com/volcengine/OpenViking）的 server 启动指引。
2. 地址不对 —— 你可以重新告诉我正确的地址。
3. 网络不通（防火墙 / VPN / 内网）—— 你确认一下网络。

也可以选择"先把配置写下来"，等服务起来就自动生效，要这么办吗？

(EN) ❌ Cannot reach BASE_URL. Likely cause:

1. Server isn't running — please ask your OpenViking admin to start it. If you own the server, follow the OpenViking official docs (https://github.com/volcengine/OpenViking) to start it. This skill does not install or run the server.
2. Wrong URL — give me the correct URL.
3. Network blocked (firewall / VPN / private network) — please verify connectivity.

Or I can save the config now (--allow-offline) so it will activate automatically once the server is up. Want me to do that?

If the user fixes the URL → back to STEP 4a. If the user wants --allow-offline → remember ALLOW_OFFLINE=true and continue to STEP 6. If the user gives up / cannot fix → stop here. Do NOT continue with a broken state.

STEP 6: Install the Plugin

The plugin can be installed two ways. Always try Path A first. Use Path B only as a backup when Path A fails because ClawHub is unavailable, rate-limited, or authentication blocks anonymous install. For version conflicts, dependency errors, or other non-registry failures, stop and show the user the error instead of switching paths silently.

Path A — Primary: openclaw plugins install (uses ClawHub)

Tell the user:

(CN) 现在开始装插件…… (EN) Installing the plugin now...

Run:

openclaw plugins install clawhub:@openviking/openclaw-plugin

Trigger fallback to Path B only if the output contains any of these strings:

• 429
• rate limit / rate-limited / Too Many Requests
• not logged in / please log in / please login / unauthorized / 401 / 403 together with clawhub
• ETIMEDOUT / ECONNRESET on a clawhub-related host
• generic message indicating the registry refused an anonymous client

Before falling back, also try the explicit registry prefix once:

openclaw plugins install clawhub:@openviking/openclaw-plugin

If the install succeeds, jump to STEP 7.

If both attempts fail with one of the fallback-eligible errors above, go to Path B. If the failure is a version conflict, missing dependency, package validation error, or another non-registry error, stop and show the last 30 lines to the user.

Path B — Backup: ov-install (bypasses ClawHub)

Tell the user:

(CN) ClawHub 现在好像被限流、不可用，或者当前账号不能安装。我改用备用路径，通过 npm 下载并部署插件包。 (EN) ClawHub looks rate-limited, unavailable, or blocked for this account. I'll use the backup path and install the plugin package from npm.

Run the installer with npx (no global install needed):

npx -y openclaw-openviking-setup-helper@latest --base-url BASE_URL [--api-key API_KEY] [--peer-role PEER_ROLE] [--peer-prefix PEER_PREFIX] [--account-id ACCOUNT_ID] [--user-id USER_ID]

Build the flag list according to what the user gave you:

• Always pass --base-url BASE_URL.
• Pass --api-key API_KEY only if API_KEY is non-empty.
• Pass --peer-role PEER_ROLE only if the user chose assistant or sender; omit it for the none default. Normalize legacy person to sender first.
• Pass --peer-prefix PEER_PREFIX only if the user gave one and PEER_ROLE=assistant.
• --account-id / --user-id only if the root-key path requires them.

ov-install will, in one shot:

1. Download the @openviking/openclaw-plugin package from npm into a temporary staging dir.
2. Copy the package into the OpenClaw extensions/ dir and install plugin dependencies.
3. Register the plugin in openclaw.json (via openclaw plugins enable or direct write).
4. Run openclaw openviking setup --json --base-url … [--api-key …] for the user.
5. Return a non-zero exit if setup needs explicit --allow-offline or --force-slot consent.

This means STEP 7 is effectively done by ov-install. After ov-install exits 0, jump straight to STEP 9 (gateway restart) and STEP 10 (verify).

If ov-install exits non-zero, capture the last 30 lines of its output, show them to the user, and stop. Don't retry blindly.

STEP 7: Configure (only on Path A — Path B did this for you)

Run the setup wizard non-interactively. Build flags from collected values:

openclaw openviking setup --base-url BASE_URL --json [--api-key API_KEY] [--peer-role PEER_ROLE] [--peer-prefix PEER_PREFIX] [--account-id ACCOUNT_ID] [--user-id USER_ID] [--allow-offline] [--force-slot]

Rules:

• --base-url BASE_URL is required under --json. Without it, the wizard prints --json requires --base-url for non-interactive mode.
• --api-key only if API_KEY is non-empty.
• --peer-role only if the user chose assistant or sender; omit it for the none default. Normalize legacy person to sender first.
• --peer-prefix only if the user gave one and --peer-role assistant is in play. Use --peer-prefix for assistant peer prefixes; the old ID-style setup flag is no longer supported.
• --account-id / --user-id only after STEP 7 root-key detection (see below).
• --allow-offline only if the user explicitly approved it in STEP 5.
• --force-slot never in the first attempt. Add only after the user confirms (see slot_blocked handling below).

Parse the JSON output

The wizard prints a single JSON object:

{
  "success": true | false,
  "action": "configured" | "existing" | "error" | "slot_blocked",
  "config": { "mode": "remote", "baseUrl": "...", "apiKey": "...", "peer_prefix": "...", "accountId": "...", "userId": "..." },
  "health": { "ok": true, "version": "...", "compatibility": "compatible" | "server_too_old" | "server_too_new" | "unknown" },
  "keyProbe": { "keyType": "user_key" | "root_key" | "no_key" | "unknown", "needsAccountId": false, "needsUserId": false, "detail": "..." },
  "slot": { "activated": true, "replaced": false, "previousOwner": "openviking" },
  "error": "..."
}

Decision matrix

Slot conflict (slot_blocked)

The error text looks like:

contextEngine slot is owned by "<other-plugin>". Config was saved but slot was NOT changed. Use --force-slot to replace.

Do NOT silently use --force-slot. Ask the user:

(CN) 你的 OpenClaw 当前 contextEngine 槽被 <other-plugin> 占着。如果用 OpenViking 替换它，<other-plugin> 就不再生效。要替换吗？ (EN) Your contextEngine slot is currently owned by <other-plugin>. Activating OpenViking will disable it. Replace?

If the user agrees, retry the same setup command with --force-slot appended. If they decline, tell them config has been saved but the slot is unchanged, and stop.

Root-key path

The error text looks like:

Root API key detected. Missing: --account-id, --user-id. Provide both to enable multi-tenant scoping.

Ask the user:

(CN) 你给的是 root 级 API Key，需要再补两个值才能用：账户 ID（accountId）和用户 ID（userId）。这两个一般是 OpenViking 服务管理员配的，不知道就问他们。 (EN) Your API key is a root key, which needs two more values: accountId and userId. Both come from your OpenViking server admin — ask them if unsure.

After collecting, retry the setup command with --account-id and --user-id appended.

STEP 8: (Reserved — done inside STEP 7 or by ov-install)

The setup wizard already wrote plugins.entries.openviking.config.* and (if successful) set plugins.slots.contextEngine = "openviking". There is no separate STEP 8 — go to STEP 9.

STEP 9: Restart the Gateway

openclaw gateway restart

If it fails, try once more with:

openclaw gateway --force

If both fail:

(CN) Gateway 没能自动重启。请你手动跑一下 openclaw gateway restart。重启完告诉我，我来验证。 (EN) Gateway didn't restart cleanly. Please run openclaw gateway restart manually, then tell me when it's done so I can verify.

Wait ~3 seconds before STEP 10.

STEP 10: Verify

openclaw openviking status --json

Expected output:

{
  "configured": true,
  "slotActive": true,
  "health": { "ok": true },
  "config": { "baseUrl": "...", "peer_prefix": "..." }
}

Decision

• configured: true and slotActive: true and health.ok: true → success. Send the success message below.
• configured: true but slotActive: false → if you previously skipped --force-slot, ask the user about replacing now and retry STEP 7 with --force-slot. Otherwise it's a gateway restart issue — back to STEP 9.
• configured: true but health.ok: false → server became unreachable. Tell the user; the config will start working as soon as the server is back.
• configured: false → setup didn't actually persist. Show full status JSON to the user and stop.

Success message

(CN) 🎉 全部搞定！OpenViking 长期记忆已经接好了。

工作方式：

• 每一轮：你的对话会被自动追加到 OpenViking session 里；自动抽取依赖阈值 commit 或 /compact
• 明确要记住时：如果你说“记住/保存/存一下”某个长期事实，我会用 memory_store 立即提交到记忆管线
• /compact 时：待提交的 session 消息会在 OpenViking 服务端被抽取成长期记忆
• 后续会话：每次回复前我都会自动搜一下相关记忆并带进上下文

想验证一下吗？可以直接说“记住我的邮箱是 test@example.com”，让我通过 memory_store 立即提交；或者先聊几句包含可记忆事实的话，再调一下 /compact。之后开新对话问“我的邮箱是？”，应该就能召回了。

(EN) 🎉 All set! OpenViking long-term memory is connected.

How it works:

• Every turn: our messages are appended to an OpenViking session; automatic extraction depends on a threshold commit or /compact
• Explicit remember requests: if you ask me to remember/save/store a long-term fact, I can use memory_store to commit it immediately
• On /compact: pending session messages are extracted into long-term memories on the OpenViking server
• Future sessions: relevant memories are auto-retrieved and injected before my replies

Want to verify? Say "remember my email is test@example.com" so I can commit it via memory_store, or tell me a few memory-worthy facts and run /compact. Then start a new chat and ask "what's my email?" — it should recall.

Part 2 — Reference

How It Works

The context-engine pipeline has three distinct stages plus one explicit write path — keep them apart, especially when telling users when memories become searchable long-term memory on the OpenViking server:

• Archive / capture (context-engine afterTurn): at the end of a user turn, the plugin appends user/assistant messages to the OpenViking session via POST /api/v1/sessions/.../messages. This is session capture only unless pending_tokens crosses commitTokenThreshold; below the threshold, no memory extraction runs yet. You'll see session message counts grow on the server, but no new files under viking://user/.../memories/.
• Memory extraction (threshold commit or /compact): memory extraction runs after a session commit. The commit can be triggered asynchronously when afterTurn crosses commitTokenThreshold, synchronously when the user invokes OpenClaw's /compact command, or explicitly by memory_store. The server-side extraction pipeline reads the archived session and writes new memories.
  • captureMode: "semantic" (default): server extraction pipeline filters all qualifying text.
  • captureMode: "keyword": only text matching trigger words (e.g. "remember", "preference") is considered.
• Auto-Recall (context-engine assemble()): before prompt context is assembled, the plugin queries OpenViking for relevant memories and injects them into context. Recall works even when there are no extracted memories yet — you just won't see anything come back.

Practical implication for testing: if you write down a short fact and immediately try to recall it without a threshold commit, /compact, or memory_store, the plugin may only retrieve it as recent session context, not as a long-term memory. To verify long-term memory cross-session deterministically, run /compact or use memory_store for the fact being tested.

Explicit long-term memory writes

Auto-capture is best-effort and commit-dependent. When the user explicitly says to remember, save, or store an important long-term fact, preference, project, or decision, the agent should call memory_store instead of waiting for ordinary auto-capture.

Use memory_store as the integration-side reliable path for durable-memory intent:

• It writes the supplied text into an OpenViking session and calls commit(wait=true).
• It complements auto-capture; it does not replace normal session capture.
• If it commits but extracts 0 memories, the explicit path has done its job. Treat that as a server-side extraction/model/configuration issue and check OpenViking logs.

Available Tools

These are the plugin tools the agent can call once installed.

memory_recall — Search Memories

Example: user asks "What programming language did I say I like?"

memory_store — Manual Store

Use this when the user explicitly asks to remember/save/store a long-term fact, preference, project, or decision.

Example: user says "Remember my email is xxx@example.com".

memory_forget — Delete Memories

Example: user says "Forget my phone number".

ov_archive_search — Search Archived Original Messages

Use when [Session History Summary] is too coarse. Try at least two keyword variants before concluding the detail is unavailable.

ov_archive_expand — Expand An Archive

add_resource — Import Resource

The agent-visible add_resource tool is disabled by default (enableAddResourceTool=false). Use manual /add-resource for resource ingestion unless you explicitly opt in. Even when enabled, never use add_resource during search, retrieval, URI reading, or search-result optimization; use ov_search and ov_read for those flows.

add_skill — Import Agent Skill

Agent Skill best practice: verify frontmatter has a precise name, a trigger-oriented description, useful tags, explicit scope boundaries, and concrete execution steps. Do not include secrets in skill content.

ov_search — Search Resources and Skills

Important: ov_search result URIs are OpenViking virtual URIs, not local file paths. If full content is needed, call ov_read with the exact viking://... URI returned by search or trace results; do not use filesystem read tools.

ov_read — Read OpenViking Content

ov_recall_trace — Query Recall Trace

Requires traceRecall=true; persisted lookup also requires traceRecallPersist=true.

Externalized Tool Result Tools

Use these when a preview contains a viking://session/<session_id>/tool-results/<tool_result_id> ref.

Configuration Schema

These are the keys under plugins.entries.openviking.config in openclaw.json. The setup wizard / ov-install sets the first few; the rest are tunables.

To change a value:

openclaw config set plugins.entries.openviking.config.<field> <value>
openclaw gateway restart

Multi-Tenant (Root API Keys)

Some OpenViking deployments use a single root API key shared across tenants. In that case the plugin needs both accountId and userId so it can scope memories correctly. The setup wizard detects this automatically and returns:

Root API key detected. Missing: --account-id, --user-id

When you see this:

1. Ask the user for both values (they come from the OpenViking admin).
2. Retry STEP 7 with both flags:

openclaw openviking setup --base-url BASE_URL --api-key API_KEY --account-id ACCOUNT_ID --user-id USER_ID --json

A user key (issued per tenant) does not need these flags.

Multi-Instance (--workdir / OPENCLAW_STATE_DIR)

If the user runs multiple OpenClaw instances (e.g. testing several agents in parallel), each has its own state dir.

To target a non-default instance:

npx -y openclaw-openviking-setup-helper@latest --workdir ~/.openclaw-second --base-url ... --api-key ...

ov-install writes a helper env file when the state dir is non-default:

• Unix: ~/.openclaw/openviking.env containing export OPENCLAW_STATE_DIR='...'
• Windows: ~/.openclaw/openviking.env.bat and .ps1 setting the same variable

Source it before running openclaw commands so they hit the correct state:

Unix:

source ~/.openclaw/openviking.env
openclaw status

Windows (PowerShell):

. "$HOME/.openclaw/openviking.env.ps1"
openclaw status

Or pass --workdir directly to each openclaw invocation (note: not all openclaw subcommands honor --workdir consistently — when in doubt, prefer the env var).

Daily Operations

# Start or restart OpenClaw gateway after config changes
openclaw gateway restart

# Check overall status
openclaw status
openclaw openviking status --json

# Read current OpenViking slot
openclaw config get plugins.slots.contextEngine

# Disable OpenViking memory (keep config, deactivate slot)
openclaw config set plugins.slots.contextEngine legacy
openclaw gateway restart

# Re-enable
openclaw config set plugins.slots.contextEngine openviking
openclaw gateway restart

Uninstall

Preferred: via OpenClaw plugin manager

openclaw plugins uninstall openviking
openclaw config set plugins.slots.contextEngine legacy
openclaw gateway restart

If installed through backup ov-install

npx -y openclaw-openviking-setup-helper@latest --uninstall

This will:

1. Stop the OpenClaw gateway.
2. Back up openclaw.json.
3. Remove all OpenViking plugin config from openclaw.json (plugins.entries.openviking, plugins.allow, plugins.installs.openviking, plugins.slots.contextEngine).
4. Move the plugin directory to disabled-extensions/ as backup.
5. Remove the helper env files (openviking.env, .bat, .ps1).

Add --base-url dummy for fully non-interactive mode (skips the confirmation prompt).

Manual

openclaw gateway stop
openclaw config set plugins.slots.contextEngine legacy
# Then edit ~/.openclaw/openclaw.json and remove openviking from
#   plugins.entries
#   plugins.allow
#   plugins.installs
openclaw gateway restart

Error Recovery Reference

Match against actual stderr / JSON error strings.

Important Rules

1. Never ask the user to run commands. You run everything via your shell tool.
2. Never skip STEP 5 (connectivity check). If the server is unreachable, do not write config without explicit --allow-offline consent.
3. Never silently use --force-slot. Slot replacement disables another plugin — always confirm with the user first.
4. Never invent values. If the user can't provide a required value, stop and tell them what to ask their admin.
5. Never claim success without STEP 10. Only after openclaw openviking status --json shows configured: true && slotActive: true && health.ok: true may you tell the user it's done.
6. Use --peer-prefix for assistant peer prefixes. The old ID-style setup flag is no longer supported by the schema.
7. For Windows, use PowerShell equivalents. Don't rely on nohup, &, mkdir -p, source, etc.
8. Switch to Path B (ov-install) only for ClawHub/rate-limit/registry availability failures. Don't use it to hide version conflicts or package validation errors.
9. Do NOT install or operate the OpenViking server. This skill assumes the server is already running. If it isn't, tell the user to contact their admin or follow the OpenViking docs.
10. Be brief and friendly in user-visible text. Save technical detail for when something actually fails.
11. Do NOT use clawhub install openviking. That installs a different thing (an AgentSkill, not the plugin).