Pika launches PikaStream 1.0 video chat for any agent

• Pika launched a beta video chat skill called PikaStream 1.0 that lets an agent appear on a live call with memory, personality, and real-time adaptation Pika launch.
• The release is not limited to Pika’s in-house avatars. Pika also published the skill on GitHub so Claude Code, OpenClaw, and other coding agents can join Google Meet calls too GitHub skill link.
• The public repo prices the pikastream-video-meeting skill at $0.50 per minute and describes extras that are more ambitious than a simple call bot: voice cloning, avatar generation, workspace-aware prompts, and post-meeting notes (Pika-Skills).
• Pika is positioning its own “AI Self” as the easiest entry point. The product site says those selves are built from a selfie, a voice recording, and personality questions, then continue to remember and evolve over time AI Self link (Pika).
• The company is calling the release a beta, warning about glitches, and routing feedback into a new Discord server Beta disclaimer.

You can browse the source, check the repo’s four CLI verbs for joining, leaving, avatar generation, and voice cloning, and see that the launch is wired specifically around Google Meet rather than a generic meeting abstraction. Pika’s own site also frames the product around persistent AI selves that “talk, post, remember, and grow” (Pika), which makes the video skill feel less like a one-off demo and more like a new interface layer.

Pika’s headline claim is simple: video chat is now a skill for any agent, not just a feature inside Pika’s own app. The launch post says the beta runs on a new real-time model called PikaStream 1.0, preserves memory and personality, and can execute agentic tasks during the call when paired with a Pika AI Self.

That “any agent” line is the part creative tool people will care about. Pika immediately paired the announcement with a GitHub release path for external agents, which turns the launch from product theater into something builders can actually wire into their own stack GitHub skill link.

The Pika-Skills repository is unusually concrete about what the system does. It exposes one skill today, pikastream-video-meeting, with this feature list:

• Join a Google Meet as a real-time avatar
• Generate an avatar image, or use a custom one
• Clone a voice from a short audio sample
• Check balance before joining, then generate a payment link if needed
• Build a conversation prompt from workspace context
• Retrieve and share notes after the meeting

The repo also publishes the command surface, which gives the launch more shape than the tweets do. There are separate commands to join a meeting, leave a meeting, generate an avatar, and clone a voice. Requirements are lightweight: Python 3.10+, a PIKA_DEV_KEY, and optional ffmpeg for voice-cloning audio conversion.

Pika is still steering users toward its own AI Self product first. The launch thread says you can ask your AI Self to join a Google Meet, while everyone else can install the GitHub skill manually GitHub skill link.

On Pika’s site, an AI Self is created by uploading a selfie, recording a voice, and answering personality questions. The homepage says the result can talk, post, remember, and grow over time. That explains why the announcement leans so hard on continuity words like memory and personality. Pika is selling the call as a live surface for an already-persistent character, not just a webcam wrapper around a model.

Pika attached two practical caveats to the launch. First, it says the feature is still in beta and will glitch. Second, the public repo makes clear that usage is metered: the only listed skill is priced at $0.50 per minute.

The company is collecting feedback through a new Discord server, and early reactions on X focused less on raw model quality than on the uncanny feeling of sending an AI into a Meet call on your behalf Early reaction User demo repost. That social response is its own product clue. Pika did not just ship another chat box, it shipped a way to put an agent on camera.