manifest
Smart LLM Router for OpenClaw. Save up to 70% by routing every request to the right model. No coding required.
- License 许可
- MIT
- 最近更新
- 2026-06-07
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- manifest
- metadata
- { "openclaw": { "homepage": "https://github.com/mnfst/manifest", "requires": { "env": [ "MANIFEST_API_KEY" ], "bins": [ "openclaw" ], "config": [ "plugins.entries.manifest.config.apiKey" ] }, "primaryEnv": "MANIFEST_API_KEY" } }
- description
- Smart LLM Router for OpenClaw. Save up to 70% by routing every request to the right model. No coding required.
03 SKILL.md 正文
Manifest — LLM Router & Observability for OpenClaw
Manifest is an OpenClaw plugin that:
- Routes every request to the most cost-effective model via a 23-dimension scoring algorithm (<2ms latency)
- Tracks costs and tokens in a real-time dashboard
- Sets limits with email alerts and hard spending caps
Source: github.com/mnfst/manifest — MIT licensed. Homepage: manifest.build
Setup (Cloud — default)
Three commands, no coding:
openclaw plugins install manifest
openclaw config set plugins.entries.manifest.config.apiKey "mnfst_YOUR_KEY"
openclaw gateway restart
Get the API key at app.manifest.build → create an account → create an agent → copy the mnfst_* key.
After restart, the plugin auto-configures:
- Adds
manifest/autoto the model allowlist (does not change your current default) - Injects the
manifestprovider into~/.openclaw/openclaw.json - Starts exporting OTLP telemetry to
app.manifest.build - Exposes three agent tools:
manifest_usage,manifest_costs,manifest_health
Dashboard at app.manifest.build. Telemetry arrives within 10-30 seconds (batched OTLP export).
Verify connection
openclaw manifest
Shows: mode, endpoint reachability, auth validity, agent name.
Configuration Changes
On plugin registration, Manifest writes to these files:
| File | Change | Reversible |
|---|---|---|
~/.openclaw/openclaw.json |
Adds models.providers.manifest provider entry; adds manifest/auto to agents.defaults.models allowlist |
Yes — openclaw plugins uninstall manifest |
~/.openclaw/agents/*/agent/auth-profiles.json |
Adds manifest:default auth profile |
Yes — uninstall removes it |
~/.openclaw/manifest/config.json |
Stores auto-generated API key (local mode only, file mode 0600) | Yes — delete ~/.openclaw/manifest/ |
~/.openclaw/manifest/manifest.db |
SQLite database (local mode only) | Yes — delete the file |
No other files are modified. The plugin does not change your current default model.
Install Provenance
openclaw plugins install manifest installs the manifest npm package.
- Source: github.com/mnfst/manifest (
packages/openclaw-plugin/) - License: MIT
- Author: MNFST Inc.
Verify before installing:
npm view manifest repository.url
npm view manifest dist.integrity
Setup (Local — offline alternative)
Use local mode only when data must never leave the machine.
openclaw plugins install manifest
openclaw config set plugins.entries.manifest.config.mode local
openclaw gateway restart
Dashboard opens at http://127.0.0.1:2099. Data stored locally in ~/.openclaw/manifest/manifest.db. No account or API key needed.
To expose over Tailscale (requires Tailscale on both devices, only accessible within your Tailnet): tailscale serve --bg 2099
What Manifest Answers
Manifest answers these questions about your OpenClaw agents — via the dashboard or directly in-conversation via agent tools:
Spending & budget
- How much have I spent today / this week / this month?
- What's my cost breakdown by model?
- Which model consumes the biggest share of my budget?
- Am I approaching my spending limit?
Token consumption
- How many tokens has my agent used (input vs. output)?
- What's my token trend compared to the previous period?
- How much cache am I reading vs. writing?
Activity & performance
- How many LLM calls has my agent made?
- How long do LLM calls take (latency)?
- Are there errors or rate limits occurring? What are the error messages?
- Which skills/tools are running and how often?
Routing intelligence
- What routing tier (simple/standard/complex/reasoning) was each request assigned?
- Why was a specific tier chosen?
- What model pricing is available across all providers?
Connectivity
- Is Manifest connected and healthy?
- Is telemetry flowing correctly?
Agent Tools
Three tools are available to the agent in-conversation:
| Tool | Trigger phrases | What it returns |
|---|---|---|
manifest_usage |
"how many tokens", "token usage", "consumption" | Total, input, output, cache-read tokens + action count for today/week/month |
manifest_costs |
"how much spent", "costs", "money burned" | Cost breakdown by model in USD for today/week/month |
manifest_health |
"is monitoring working", "connectivity test" | Endpoint reachable, auth valid, agent name, status |
Each accepts a period parameter: "today", "week", or "month".
All three tools are read-only — they query the agent's own usage data and never send message content.
LLM Routing
When the model is set to manifest/auto, the router scores each conversation across 23 dimensions and assigns one of 4 tiers:
| Tier | Use case | Examples |
|---|---|---|
| Simple | Greetings, confirmations, short lookups | "hi", "yes", "what time is it" |
| Standard | General tasks, balanced quality/cost | "summarize this", "write a test" |
| Complex | Multi-step reasoning, nuanced analysis | "compare these architectures", "debug this stack trace" |
| Reasoning | Formal logic, proofs, critical planning | "prove this theorem", "design a migration strategy" |
Each tier maps to a model. Default models are auto-assigned per provider, but overridable in the dashboard under Routing.
Short-circuit rules:
- Messages <50 chars with no tools → Simple
- Formal logic keywords → Reasoning
- Tools present → floor at Standard
- Context >50k tokens → floor at Complex
Dashboard Pages
| Page | What it shows |
|---|---|
| Workspace | All connected agents as cards with sparkline activity charts |
| Overview | Per-agent cost, tokens, messages with trend badges and time-series charts |
| Messages | Full paginated message log with filters (status, model, cost range) |
| Routing | 4-tier model config, provider connections, enable/disable routing |
| Limits | Email alerts and hard spending caps (tokens or cost, per hour/day/week/month) |
| Settings | Agent rename, delete, OTLP key management |
| Model Prices | Sortable table of 300+ model prices across all providers |
Supported Providers
Anthropic, OpenAI, Google Gemini, DeepSeek, xAI, Mistral AI, Qwen, MiniMax, Kimi, Amazon Nova, Z.ai, OpenRouter, Ollama. 300+ models total.
Uninstall
openclaw plugins uninstall manifest
openclaw gateway restart
This removes the plugin, provider config, and auth profiles. After uninstalling, manifest/auto is no longer available. If any agent uses it, switch to another model.
Troubleshooting
Telemetry not appearing: The gateway batches OTLP data every 10-30 seconds. Wait, then check openclaw manifest for connection status.
Auth errors in cloud mode: Verify the API key starts with mnfst_ and matches the key in the dashboard under Settings → Agent setup.
Port conflict in local mode: If port 2099 is busy, the plugin checks if the existing process is Manifest and reuses it. To change the port: openclaw config set plugins.entries.manifest.config.port <PORT>.
Plugin conflicts: Manifest conflicts with the built-in diagnostics-otel plugin. Disable it before enabling Manifest.
After backend restart: Always restart the gateway too (openclaw gateway restart) — the OTLP pipeline doesn't auto-reconnect.
Privacy
External Endpoints
| Endpoint | When | Data Sent |
|---|---|---|
{endpoint}/v1/traces |
Every LLM call (batched 10-30s) | OTLP spans (see fields below) |
{endpoint}/v1/metrics |
Every 10-30s | Counters: request count, token totals, tool call counts — grouped by model/provider |
{endpoint}/api/v1/routing/resolve |
Only when model is manifest/auto |
Last 10 non-system/non-developer messages ({role, content} only) |
https://eu.i.posthog.com |
On plugin register | Hashed machine ID, OS, Node version, plugin version, mode. No PII. Opt out: MANIFEST_TELEMETRY_OPTOUT=1 |
OTLP Span Fields
Exhaustive list of attributes sent per span:
openclaw.session.key— opaque session identifieropenclaw.message.channel— message channel namegen_ai.request.model— model name (e.g. "claude-sonnet-4-20250514")gen_ai.system— provider name (e.g. "anthropic")gen_ai.usage.input_tokens— integergen_ai.usage.output_tokens— integergen_ai.usage.cache_read_input_tokens— integergen_ai.usage.cache_creation_input_tokens— integeropenclaw.agent.name— agent identifiertool.name— tool name stringtool.success— booleanmanifest.routing.tier— routing tier (if routed)manifest.routing.reason— routing reason (if routed)- Error status: agent errors truncated to 500 chars; tool errors include
event.error.messageuntruncated
Not collected: user prompts, assistant responses, tool input/output arguments, file contents, or any message body.
Routing Data
- Only active when model is
manifest/auto - Excludes messages with
role: "system"orrole: "developer" - Sends only
{role, content}— all other message properties are stripped (routing.ts:77) - 3-second timeout, non-blocking
- To avoid sending content: disable routing in the dashboard or set a fixed model
Credential Storage
- Cloud mode: API key stored in
~/.openclaw/openclaw.jsonunderplugins.entries.manifest.config.apiKey(managed by OpenClaw's standard plugin config) - Local mode: auto-generated key stored in
~/.openclaw/manifest/config.jsonwith file mode0600
Local Mode
All data stays on your machine. No external calls except optional PostHog analytics (opt out with MANIFEST_TELEMETRY_OPTOUT=1).
04 同仓库更多 Skills
crowd-prompting
A marketplace where AI agents improve prompts, system instructions, tool descriptions, and other text-based content with domain expertise from real-world operations — and earn tokens for valuable contributions.
ceaser
Interact with the Ceaser privacy protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- shield, unshield, note management, and protocol queries. All ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). Ceaser is a privacy-preserving ETH wrapper using Noir/UltraHonk zero-knowledge proofs on Base L2 (chain ID 8453).
ceaser-send
Fully automated private ETH transfer via Ceaser Protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- all ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). WARNING -- creates observable on-chain links between user wallet and hot wallet (see Privacy Warning). Generates an ephemeral hot wallet with BIP-39 mnemonic recovery, user funds it, agent signs and broadcasts the Shield TX automatically. Exactly one manual step required (funding the hot wallet with ETH). Uses Noir/UltraHonk zero-knowledge proofs.
dilemm-ai
Compete in DilemmAI, the prisoner's dilemma AI arena at dilemm.ai. Use when an OpenClaw agent wants to create an account, design and submit strategy prompts for their bot, enter matchmaking, analyze historical matches and leaderboards, and iterate on bot strategy. Requires headless browser automation (no API available yet).
moldium
Post and manage content on the Moldium blog platform. Triggered by "post to Moldium", "write a blog post", "publish an article", etc.
arxiv-paper-reviews
Interact with arXiv Crawler API to fetch papers, read reviews, submit comments, search papers, and import papers. Use when working with arXiv papers, fetching paper lists by date/category/interest, viewing paper details with comments, submitting paper reviews, searching papers by title, or importing papers from arXiv URLs via API at http://weakaccept.top:8000/.