ebb-ai
v0.11.0 · operator preview

Defer AI work to balance the grid — cheaper, faster, lower-carbon.

US AI compute is projected to reach 6.7–12 % of national grid load by 2028 (DOE 2024). ebb-ai is an open-source MCP scheduler that defers non-urgent LLM tasks to off-peak hours — 50 % cheaper via Anthropic/OpenAI Batch APIs, faster during providers' expanded off-peak capacity, and 40–70 % lower carbon. Per-task receipts to a local SQLite ledger. Apache-2.0.

install — 30 seconds

One line. Any host.

Paste the prompt below into your agent. It detects the host (Claude Code, Cursor, Claude Desktop, Windsurf, Continue, Cline, Zed, Goose, OpenClaw, or any generic MCP host) and runs the right install path for you.

agent prompt
Please install ebb-ai — open-source carbon-aware MCP scheduler, Apache-2.0.

Instructions: https://www.ebb-ai.com/install.md

Read that file, then run the install path for whichever host
you're in and confirm it worked.

install it yourself

Pick your host below and copy the command. Default is the universal MCP install; switch the dropdown for host-specific configs or to use the library directly.

Universal — add ebb-mcp to any host that speaks Model Context Protocol.

bash
npx -y @ebb-ai/mcp

then Wire this into your host's MCP config (the snippets for specific hosts are in the dropdown).

Once installed, your assistant can call /ebb-ai:defer "summarize this" --by tomorrow 6pm and the task lands at the cleanest grid hour inside the deadline. See all commands →

how it works

One scheduled task, four parallel wins.

A task arrives while the grid is at peak — dirty, strained, expensive. ebb-ai rides it down the carbon curve and dispatches it at the clean off-peak trough. From that one dispatch, four wins flow out at once — carbon, cost, latency and grid load — because they all track the same thing: grid demand.

peak demand · dirtyoff-peak trough · cleanPEAK · dirtyCLEAN · off-peakAI task

Carbon

−80%

vs dispatching at peak

Cost

−50%

Batch-API window

Latency

Faster

off-peak capacity

Grid load

−83%

demand time-shifted

Illustrative — the loop shows the mechanism, not live data. A deferrable task waits through the dirty, strained peak and is dispatched at the off-peak trough inside its deadline. Carbon, cost and latency move together because they all track grid demand; exact figures vary per region per day.

dashboard

Live data + tools.

grid regions
31
live carbon feeds
4
carbon forecast
72h
intensity bands
5
Carbon-aware scheduling for AI workflows