Livepeer 2.0: The Open Video Agent Platform
Livepeer 2.0 is in motion. Here's what's changing in the network and the protocol, and what's still to come.
Agents are taking over the video production line. The tools to generate, edit, and transform video are now embedded in the daily workflow of millions of creators, and yet almost all of them sit behind a closed platform and a tight rate limit.
Livepeer's open infrastructure is the counterforce. We believe that video infrastructure, and the creativity it enables, should be open all the way down.
The next chapter for Livepeer
Two weeks ago, Livepeer co-founder Doug Petkanics laid out A Path to Livepeer 2.0. After nine years of critical engineering work to build a decentralized GPU network, this ambitious set of upgrades across the Livepeer Network and Protocol enable Livepeer to go to market for the first time without relying on third-party products or applications. This next chapter for Livepeer gives AI agents a truly open network of AI models, tools, and compute for creating and transforming video and media, while still allowing for open experimentation on the network. The three primary changes wrapped in the next chapter for Livepeer include:
Connecting the protocol to usage
The proposed protocol updates include a Burn-Mint Equilibrium, a stake-elected validator set, and fixed per-node bonds in place of the 100-orchestrator cap. All of these design mechanisms intend to point at one outcome: where LPT rewards flow to nodes performing and securing quality work connected directly to demand.
This is where we need our delegators, node operators, and contributors to serve as thoughtful design partners to solidify our approach to the most critical protocol design questions.
Growing beyond a GPU network
Livepeer's node operators historically competed on one thing: hardware. The node operator role was relatively straightforward, as it followed a core loop of 1) bring GPUs, 2) run transcoding or inference, and 3) earn fees. That loop breaks under the complexity of AI, where the network now needs hundreds of models available on demand on top of performant compute service providers.
With the network changes outlined in A Path to Livepeer 2.0, nodes will serve open models locally, connect APIs, provide compute, and run non-GPU media capabilities like layering audio, clipping, captions, resolution changes, overlays, object-aware augmentation, and so much more. The unit of supply is no longer a GPU—it's compute. This opens the door to CPU-only nodes, API specialist nodes, and operators who compete on breadth and quality of service instead of just silicon.
Bringing an agent to market
The final component of this next chapter for Livepeer is making the network reachable through an MCP: a custom connector that any agent harness can call that supercharges your agent with video and moving image rendering capabilities. The Livepeer Network answers on the other side with access to the world's AI models, performant compute services, and a growing library of video transformation skills and playbooks across generating, editing, and augmenting footage, all powered by a global network of independent compute providers. More on the agent itself very soon.
The new, open Livepeer stack
- The Livepeer 2.0 Protocol is the stake-based protocol that coordinates the network, powered by LPT.
- The Livepeer 2.0 Network is a global marketplace of independent compute providers running the world's AI models.
- The Livepeer Agent transforms your prompt into a finished video—planning the painstaking editing steps, orchestrating across the best models, and settling for the work as it goes, all from inside the agent you already use.
Come build it
Read the initial roadmap and the 2.0 vision on the Livepeer Forum. If you run an orchestrator, delegate, or hold LPT, the community design sessions are where we'll explore protocol upgrades. The details and schedule are in Discord.
This is the first of many exciting updates to come—stay tuned for more next week, where we'll have something new to get your hands on.