GBrain adds voice agents

GBrain now lets people talk to their AI agents by voice, adding browser calls over Web Real-Time Communication and phone calls through Twilio in version 0.8.0. (github.com) Voice agents are software that stay in a live conversation, handling speech, interruptions, and replies in real time instead of waiting for typed prompts. OpenAI’s voice agent documentation says browser apps connect over Web Real-Time Communication using short-lived client tokens, while Twilio handles phone connectivity on the telephony side. (openai.github.io; twilio.com) Garry Tan’s open-source GBrain project is a personal knowledge system built on PostgreSQL, pgvector, and hybrid search for markdown files and agent workflows. The repository README says it is designed to make scattered notes and documents searchable by both keywords and meaning. (github.com) The new voice layer plugs into that memory system rather than replacing it. GBrain’s Twilio recipe says phone calls can create brain pages through a pipeline that connects Twilio, OpenAI Realtime, and GBrain’s Model Context Protocol tools for search, reads, and writes. (github.com) That means a spoken call can become structured notes inside the same knowledge base the agent already uses for search and follow-up work. The Twilio recipe says post-call pages are written back into the brain, turning conversations into stored records the agent can query later. (github.com) The timing fits GBrain’s recent push toward more automated agent workflows. Its upgrade guide says the software already checks for new versions, asks users for permission to upgrade, and runs migrations so new capabilities actually work in the user’s environment. (github.com) The repository also shows how fast the project is moving. GitHub lists a commit from April 12, 2026 described as “Voice v0.8.0 + feature discovery + Edge Function removal,” and the project had grown to about 4,700 stars at the time the page was crawled. (github.com) GBrain is still shipping more through commits and docs than through formal packaged releases. GitHub’s releases page showed no published releases when it was crawled, even as the repository and recipes documented the new voice workflow. (github.com; github.com) The result is a broader front door for the same idea: instead of only typing to an agent that knows your files, GBrain now supports talking to one in a browser session or over a phone line. (github.com; github.com)