Skip to content

Install Footprint Guide

wasmagent packages are composable — install only what you need.


Package footprint table

PackageInstall sizeEdge (Cloudflare Workers)Notes
@wasmagent/corelightweightNo native deps; tree-shake aggressively
@wasmagent/mcp-firewalllightweightPure TypeScript; no native deps
@wasmagent/aeplightweightPure TypeScript; Zod only
@wasmagent/mcp-policylightweightThin wrapper over mcp-firewall
@wasmagent/mcp-attestationlightweightPure TypeScript
@wasmagent/mcp-gatewaylightweightPure TypeScript
@wasmagent/compliancelightweightPure TypeScript
@wasmagent/capability-compilerlightweightPure TypeScript
@wasmagent/otel-exporterlightweightHTTP only; no native deps
@wasmagent/evals-runnerlightweightPure TypeScript
@wasmagent/kernel-quickjsmedium (~1 MB)WASM binary included
@wasmagent/kernel-wasmtimemedium❌ Node.js onlyRequires Node WebAssembly API
@wasmagent/kernel-pyodideheavy (~8 MB+)CPython-in-WASM; too large for Workers
@wasmagent/kernel-remotelightweightHTTP only; connects to E2B/Cloudflare
@wasmagent/tools-browserheavy❌ Node.js onlyRequires Playwright
@wasmagent/tools-ragmedium⚠️ partialVector DB connectors vary by backend
@wasmagent/tools-weblightweightHTTP only
@wasmagent/model-localheavy❌ Node.js onlynode-llama-cpp native bindings
@wasmagent/model-anthropiclightweightThin SDK wrapper
@wasmagent/model-openailightweightThin SDK wrapper
@wasmagent/model-deepseeklightweightThin SDK wrapper
@wasmagent/model-doubaolightweightThin SDK wrapper
@wasmagent/model-qwenlightweightThin SDK wrapper
@wasmagent/model-zhipulightweightThin SDK wrapper
@wasmagent/model-minimaxlightweightThin SDK wrapper
@wasmagent/model-moonshotlightweightThin SDK wrapper
@wasmagent/aisdklightweightai (Vercel AI SDK) is a peer dep
@wasmagent/mastra-sandboxlightweight@mastra/core is a peer dep
@wasmagent/openai-agentslightweight@openai/agents is a peer dep
@wasmagent/claude-agent-sdklightweight@anthropic-ai/claude-agent-sdk is a peer dep
@wasmagent/reactlightweightreact is a peer dep

Edge compatibility detail

Packages marked ✅ work in Cloudflare Workers with no special configuration. Packages marked ❌ require a Node.js environment (local, server, or CI). @wasmagent/kernel-remote is the recommended isolation layer for edge/serverless deployments.


Optional peer dependencies

Install these only when you use the corresponding adapter:

PackageOptional peerWhen you need it
@wasmagent/aisdkaiUsing the Vercel AI SDK integration
@wasmagent/openai-agents@openai/agentsUsing the OpenAI Agents JS SDK
@wasmagent/claude-agent-sdk@anthropic-ai/claude-agent-sdkUsing the Anthropic Claude Agent SDK
@wasmagent/mastra-sandbox@mastra/coreUsing Mastra as the agent framework
@wasmagent/tools-browserplaywrightBrowser automation tools
@wasmagent/model-localnode-llama-cppRunning local LLMs without outbound traffic

Quickstart install footprints

Trust Pack (minimal — evidence layer only)

bash
npm add @wasmagent/mcp-firewall @wasmagent/aep

~50 KB installed. Covers: tool protection, evidence emission, CI gate.

Runtime (with sandbox execution)

bash
npm add @wasmagent/core @wasmagent/kernel-quickjs

~1.3 MB installed (includes WASM binary). Covers: sandboxed code execution, tool calling agents.

Full stack (runtime + evidence + framework adapter)

bash
npm add @wasmagent/core @wasmagent/kernel-quickjs @wasmagent/mcp-firewall @wasmagent/aep
npm add @wasmagent/aisdk   # or openai-agents / claude-agent-sdk / mastra-sandbox

~1.5 MB + framework peer deps. Covers: sandboxed execution, evidence, firewall, framework integration.

Compliance research

bash
npm add @wasmagent/compliance @wasmagent/aep @wasmagent/evals-runner
pip install evomerge

Covers: constraint verification, evidence export, training data generation.


Model SDK install cost

All model adapters are thin wrappers. The actual SDK is installed once as a direct dep:

bash
npm add @anthropic-ai/sdk        # for @wasmagent/model-anthropic
npm add openai                   # for @wasmagent/model-openai
# etc.

@wasmagent/model-local with node-llama-cpp downloads model weights separately (not part of npm install). Model weights are typically 1–8 GB and are cached in ~/.cache/node-llama-cpp.

Released under the Apache-2.0 License.