Command reference

Every command below shows the CLI form and the SDK method it calls. Both authenticate the same key the same way, so a call that works one way works the other — pick whichever fits the code you are writing.

Confirms the key resolves to a workspace and reports who owns it. Neither command changes anything, and neither costs credits.

eddie login
# Signed in as devops@studio.com (workspace a1b2c3…).

eddie whoami
# devops@studio.com	a1b2c3…

const { email, workspaceId } = await eddie.whoami();

Returns

email and workspaceId as strings.

Fails when

The key is missing, malformed, or revoked.

Creates a project, uploads and imports the local files you name, and waits for analysis to finish. Returns the promptable share id — the id every later command takes.

eddie new --import ./footage/*.mov [--name "Client Interview"]
# (progress on stderr)
# a1b2c3d4e5f6…    ← the share id, on stdout

const { shareId } = await eddie.createAndImport(
  { name: "Client Interview", files: [{ path: "./footage/a.mov" }, { path: "./footage/b.mov" }] },
  { onProgress: (p) => console.error(p.stage, p.file ?? "", p.detail ?? "") },
);

Takes

One or more local file paths, an optional project name, and an optional mode — only rough_cut (the default) is wired.

Returns

{ projectId, shareId }. Use shareId for every later command; projectId is internal bookkeeping.

Duration

Blocks until analysis finishes — uploads run first, then transcription and content analysis. Minutes, for real footage.

Cost

The usual import rate. See how credits and billing work.

Adds local files to a project that already exists, and waits for the new media to finish analysis. The project stays the same share id; nothing about media already in it changes.

eddie import "$id" ./more-footage/*.mov

await eddie.addMedia("$id", [{ path: "./more-footage/c.mov" }]);

Takes

A share id, plus one or more local file paths.

Returns

{ projectId, shareId } — the same shareId you passed in.

Prompts Eddie and streams the run as it happens: activity as the model works, the reply as it is written, and a final status once the run ends.

eddie prompt "$id" "cut a 90s teaser, punchy captions"
# (activity on stderr, the reply streams to stdout)

for await (const event of eddie.prompt(shareId, "cut a 90s teaser")) {
  if (event.type === "reply") process.stdout.write(event.text);
  if (event.type === "activity") console.error("·", event.text);
  if (event.type === "done") console.error("run", event.status);
}

Takes

A share id and the prompt text.

Yields

activity, reply and notice events as the run progresses, then one done event with a status: done, failed, cancelled, queued, held or refused.

held or queued

The run has not started analysing yet — held is the import-readiness lock, queued means capacity. Neither is a failure; the CLI prints the notice and exits.

Renders an edit to MP4 and returns a download URL.

eddie export "$id" "Teaser" --format mp4 [--resolution 1080p|source]
# https://cdn.heyeddie.ai/renders/teaser….mp4

const { url } = await eddie.export(shareId, "Teaser", "mp4", "1080p");

Takes

A share id, the edit’s name exactly as it appears in the project, and optionally a resolution — 1080p (default) or source.

Returns

{ url } — the finished MP4’s download link.

Only format

mp4. NLE interchange formats (Premiere, Resolve, FCPXML, OTIO, EDL) are not wired against this client.

• Every other export format. Only MP4 renders today. A workflow that needs a Premiere or Resolve project has to go through the Eddie web application for now.
• Project modes besides Chat & Edit. new always creates a rough_cut project.
• Reading a project back — transcripts, sources, an edit’s contents. This client is write-oriented: create, import, prompt, export. Reading a project’s detail back out is the MCP surface’s job — see the tool reference.

None of these are permanent limits of the product. They are the limits of what this client wires up so far.