Livepeer

Livepeer opened early signups for the Livepeer Agent, its AI agent product, via earlyaccess.livepeer.org. The announcement frames an agent as five parts — a model, instructions, tools, memory, and a loop that plans, acts, and checks its work — with tools increasingly connecting through MCP, and the open-source layer covering images, video, and voice.

• Livepeer has continued pushing signups, pairing them with a "normal day at the Livepeer office" video clip, an explainer of how AI video credits work — credits stand in for GPU time, and five seconds of video can be 120 pictures — and a breakdown of where open video models like Wan and LTX run: your own GPU, a rented cloud GPU, or an open network that bills per job.
• The team also pairs the signups with the question of what it takes to run a live world model, where Genie 3 generates a new frame every 42 ms.
• The Agent routes media generation through Livepeer's open network, giving AI clients access to video, image, audio, 3D, and editing tools via MCP.

Livepeer shared how track owners can now turn completed work into a published post in 5 to 10 minutes, demoed with @0xcr33pt0. The workflow points an agent at a template, an AI check fills in the gaps, and the post goes out on the blog, X, and the newsletter with the owner's name attached.

The clip frames the agent as a publishing assistant for track owners, letting them surface shipped work without writing posts manually — part of Livepeer's broader push to get its agent products used both for media compute and internal workflow.

Livepeer officially published its 2.0 litepaper, written by founder Doug Petkanics, proposing a protocol redesign around AI media compute rather than video transcoding alone. The team paired the release with a two-minute explainer video and an X Space walking LPT holders, Orchestrators, and builders through the proposal, framing the document as a concrete starting point after months of community discussion, not the final spec.

In follow-up posts, Livepeer positioned the redesign in the context of an open-source LLM debate and the economics of running that compute, explaining that AI APIs price work in three common units — per token, per second or hour, and per request — and linking readers to the litepaper. A companion walkthrough of the litepaper maps the open request path — from "a tap to a GPU, a Livepeer request makes four open stops." The team has continued to drive the document alongside a video aimed at those considering self-hosting AI, pointing audiences back to the 2.0 litepaper as the reference.

Livepeer highlighted MangaFable, a community-built project by @harendrashakya_ that turns a story into manga, animation, and live-action shots. It is built with the Livepeer Agent MCP and backed by an early SPE micro-grant from @GeoDeFine.

In a social update from the Watercooler call, Livepeer's AI Agent SPE engineer @GeoDeFine shared progress on treasury diversification, fixed Embody bugs, and offered a first look at the M3 agent workspace.

• Agent telemetry via OpenTelemetry and Arize Phoenix
• Evals and datasets served over MCP
• Support for many agent runtimes in a single interface

The update is separate from Livepeer's Live Runner rollout and 2.0 litepaper discussions, and reads as a status report on the AI agent workstream rather than a product launch.

Livepeer rolled out Live Runner, a product where a Livepeer orchestrator hosts your app as-is — from a web service to live video — with the Livepeer-specific portion reduced to a few clearly marked lines of code. The announcement demonstrates running an app directly on an orchestrator rather than on centralized infrastructure.

• An explainer video walks through deployment.
• In replies, Livepeer pointed readers to the 2.0 litepaper alongside the Live Runner demo.

Livepeer is rolling out a system where every AI inference job is quoted before it runs and settled after on the network. The company explains that open models themselves cost nothing; what users pay for is the GPU compute time metered by an open network of independent operators.

Livepeer's network already supports a library of 130+ frontier and open models. Early access is available via the link in the announcement.

Livepeer announced that users with a GitHub account can get $5 of network credit per month, free, through a payments integration called Payment house. The feature lets people use the Livepeer Network for video processing without holding crypto.

The payments integration was built by eliteprox, a partner developer. Users connect via GitHub to receive the recurring monthly credit and pay for Livepeer services through traditional payment rails instead of tokens.