Sume

Agent-oriented avatar and video generation API for agents that need callable media jobs, status polling, and retrievable production media instead of a destination-only editor.

Sume describes itself as "video generation APIs redesigned for agents." The current public surface now exposes Sume Avatar 1.0, Avatar Video 1.0, API docs, and a CLI/docs path for agent workflows. Treat the June bookmark as the launch signal and the July docs as the current authority. Source: X/@davidim, 2026-06-24; Source: Sume docs, checked 2026-07-02

Routing Summary

Use Sume when evaluating agent-callable video generation APIs, especially avatar/talking-head content where the output should become a job result with durable media URLs. Route with HyperFrames and Palmier for media-agent workflows. Do not infer provider quality from the launch video; verify API docs, billing, job status behavior, and content-policy constraints before implementation.

Current Surface

The public API is server-side and workspace-scoped. The current docs describe:

  • sume/avatar/v1.0 for avatar generation;
  • sume/avatar-video/v1.0 for avatar video jobs;
  • API domains under api.sume.com, app.sume.com, media.sume.com, and sume.com;
  • durable job lifecycle states: queued, processing, completed, failed, canceled;
  • first-party Sume media URLs for results, not raw provider URLs;
  • usage, balance, catalog, asset, and model-run endpoints.

The CLI docs are agent-oriented but still source-first: they describe account verification, catalog discovery, jobs, input assets, MCP tools, local diagnostics, and future package publication. Until there is a stable package release, use the docs/source route rather than assuming an installable CLI package. Source: Sume CLI docs, checked 2026-07-02

Agent Safety and Job Model

Sume's agent docs use the right side-effect boundary: read-only inspection first, confirmation before writes, and stronger confirmation before paid generation. For automation, use --agent --json style output where available so account metadata and signed URLs are redacted or summarized. Paid requests need idempotency keys and polling/backoff instead of blind retries, because a timeout after submit is not proof that the paid job did not start. Source: Sume agent skills docs, checked 2026-07-02; Source: Jobs and results docs, checked 2026-07-02

Why It Matters

Video generation tools are moving from human-operated web apps toward API-first components that agents can call as part of a larger workflow. Sume belongs near HyperFrames and Palmier: media generation as a callable primitive rather than a separate destination app.

X Artifact

The reviewed local thumbnail and video contact sheet are launch-positioning evidence: founder/CEO messaging, "for CMOs" copy, and "best video API" claims. Preserve them as artifact context, but route implementation decisions through the docs/API surface above.


Timeline