Pika launches PikaStream 1.0 video chat skill for Google Meet and any agent

• Pika's launch thread says PikaStream 1.0 is a beta video chat skill that lets an AI join Google Meet with a face, a voice, memory, and live adaptability.
• The open Pika-Skills repo, which Pika linked from the launch, turns that into an installable skill for third-party agents such as Claude Code and OpenClaw.
• The repo spells out more than the tweet does: it lists voice cloning, avatar generation, automatic billing, context-aware prompts, and post-meeting notes, all in one pikastream-video-meeting package, according to the primary announcement.
• Pika is pitching the feature hardest around its own AI Self product, where the follow-up thread says the avatar can execute agentic tasks during the call, while the original thread also flags the release as glitchy beta software.

You can browse the open-source skill repo, see where Pika sells developer keys, and compare that with Pika's broader AI Self product page, which frames these agents as persistent personas built from a selfie, a voice sample, and a few onboarding questions.

Pika's core claim is simple: drop a Google Meet link into an agent workflow and the agent joins as a live avatar. The launch video shows the company framing that as the first video chat skill for any agent, not just for Pika's own stack.

That cross-agent angle is what got people reacting. One early repost called out Claude, OpenClaw, and Hermes by name, while another focused on the strangeness of moving from text chat to face-to-face agent calls.

The Pika-Skills repository is where the launch gets concrete. It describes skills as self-contained folders with a SKILL.md file, scripts, and dependencies that an agent can detect automatically once installed.

For pikastream-video-meeting, the repo lists six specific capabilities:

1. Real-time avatar joining via PikaStreaming.
2. Voice cloning from a short audio recording.
3. Avatar generation through OpenAI image models or a user-supplied image.
4. Automatic balance checks and payment-link creation.
5. Context-aware conversation built from workspace identity, recent activity, and known people.
6. Post-meeting note retrieval after the bot leaves.

The same README also exposes the command surface: join, leave, generate-avatar, and clone-voice. That makes this feel less like a flashy demo and more like a packaged agent integration.

Pika keeps pairing the general-purpose skill with its own AI Self product. On Pika's homepage, the company describes those AI Selves as agents that talk, post, remember, and grow, created from a selfie, a recorded voice, and a lightweight personality setup.

That framing matters because Pika's thread says AI Selves can do agentic tasks during the call, not just hold a conversation. User reactions and demo reposts leaned into that uncanny part first: the weirdness of meeting a live video version of yourself.

The repo adds the operational details missing from the launch clip. pikastream-video-meeting is priced at $0.50 per minute, requires a Pika developer key from the dev portal, and expects Python 3.10 or newer. ffmpeg is optional, but only for voice-cloning audio conversion.

Pika also says the skill checks balance before joining and will generate a payment link if credits are low. That detail, plus the beta caveat in the thread, puts the release somewhere between an SDK primitive and a consumer-facing character product.