Lambda Fluid is an Effect-based reconstruction of the core idea behind fluid serverless compute: keep a Lambda invocation alive as a worker, send multiple logical HTTP requests through one outbound tunnel, and let the Node.js event loop execute another request while earlier requests await I/O.
This is an OSS prototype and developer tool, not a hosted multi-tenant service. One orchestrator manages one application and one compatible worker pool.
Install and verify:
pnpm install
pnpm check
pnpm testNo cloud credentials are needed. These finite commands exercise the client, HTTP orchestrator, adaptive fleet, TCP tunnels, workers, and streamed responses.
pnpm bench examples/01-order.ts
pnpm bench examples/02-three-endpoints.tsThis bundles the chosen example handler into an Alchemy-managed Lambda, starts a temporary public worker ingress, and routes the same 30-request workload through real Lambda workers.
pnpm infra:login --profile builder
pnpm demo:local-aws examples/01-order.ts --profile builder
pnpm demo:local-aws examples/02-three-endpoints.ts --profile builderMode 2 requires cloudflared, creates or updates AWS resources, and can incur
charges. Read the setup and cleanup guide before running it.
Start with the documentation navigator. It includes:
- a five-minute evaluation path;
- architecture and request-flow diagrams;
- both supported execution modes;
- a package-by-package source tour;
- protocol, concurrency, routing, and lifecycle explanations;
- benchmark interpretation;
- an explicit implemented-versus-incomplete matrix.
The shortest source-reading path is:
examples/01-order.ts— ordinary application code.packages/orchestrator/src/control-plane.ts— routing and response correlation.packages/worker/src/runtime.ts— concurrent logical jobs.packages/tunnel/src/channel.ts— bidirectional framed transport.packages/aws-runtime/src/lambda-handler.ts— Lambda adapter.
The supported evaluation surface is deliberately limited to:
- fully local end-to-end execution;
- a local orchestrator controlling AWS Lambda workers.
research/ preserves the reasoning and source research that led here; docs/
describes what the checked-in implementation does now.