Your meeting
is the prompt.
Reverbic records the standup, distills a spec — goal, requirements, acceptance criteria, verified against the transcript — and hands it to Claude Code, Cursor, or Codex. As a command, an MCP tool call, or an open pull request.
Free plan: 5 distills + 3 hrs transcription / month · No credit card
Record anywhere. No bot in the room.
Record bot-free on desktop or in the browser, capture on iPhone or Apple Watch, or import Zoom cloud recordings automatically — flexible ways to capture meetings anywhere, without a "Recording Bot" tile announcing you to the whole call.
The desktop app hears the meeting. The meeting never sees it.
Native system-audio capture on macOS and Windows records any call from the menu bar — Zoom, Meet, Teams, whatever plays through your machine. One global shortcut, no meeting invite, no bot participant, no awkward "who's Reverbic Notetaker?".
- System-audio loopback, not a bot in the participant list
- Global shortcut to start from anywhere
- Slack Huddles auto-record when connected
- Transcripts sync to the same workspace as every other capture
In the browser
Record or upload from any machine — live transcript included.
On your iPhone
Hallway conversations and on-site meetings, synced when you're back.
From your wrist
Tap the complication, talk, done — the watch hands off to your phone.
Summaries and action items you can reshape, not just read
Every meeting gets an intelligent pass: summary, key points, action items, and decisions. Then it stays workable — regenerate, edit, and cut it the way your team assigns work.
- Action items extracted automatically, with owners where the transcript names them
- Per-topic snippets: cut a meeting into assignable pieces instead of one long recap
- Decisions tracked across meetings, searchable later
- Keep the library organized — search, sort, and share meetings across the workspace
Ask the meeting. Keep the answer.
"What did we agree on the migration order?" Ask any meeting a question and get an answer grounded in the transcript, with timestamp citations — or an honest "that wasn't discussed." Save the good answers to the meeting so the next person doesn't have to ask.
Watch a standup become a branch
Four stages. Click through them — each panel is the actual artifact the product produces at that step, including the one nobody else does: verification.
Raw transcript, vocabulary-primed
Standup · Jul 10
42:17From mind map to merged PR.
Talk through the build and watch it take shape: ideas cluster into a live board, the board grounds itself in your GitHub repo, and Claude drafts a reviewable pull request. Preview-only until you confirm — then it pushes the branch and opens the PR for you.
Mind map
As the conversation happens, speech clusters into a live board of typed nodes — functions, tables, endpoints — that you can drag, merge, and edit in real time.
GitHub exploration
Atmos reads your AGENTS.md, CLAUDE.md, .cursorrules and searches the repo for existing helpers and conventions it should build on — not around.
Code generation
Claude drafts the change grounded in what it found: reuse-aware, convention-matched. Every proposed file is previewed before anything is written.
Auto-push · PR
One confirm pushes the branch and opens the pull request on GitHub. Review it, request changes, or merge — it's a normal PR.
Otter, Fireflies, and Fathom remember your meetings.
They can't build from them.
AI notetakers end at the summary — someone still translates it into a ticket, a prompt, a branch. Reverbic starts where they stop: the meeting leaves as work your coding agent can execute.
Competitor column describes the AI-notetaker category as positioned on their own sites, July 2026.
Stop exporting. Let the agent ask.
ChatGPT, Claude, and any MCP-capable tool can securely access your meeting knowledge through Reverbic's MCP server at reverbic.ai/api/mcp — deeper analysis, smarter workflows, zero copy-paste. Mint a scoped key, drop in the config, and your agent queries meetings, action items, and decisions mid-task.
{
"mcpServers": {
"reverbic": {
"command": "npx",
"args": [
"-y", "mcp-remote",
"https://reverbic.ai/api/mcp",
"--header", "Authorization:Bearer rvb_..."
]
}
}
}It works both directions: register the MCP servers your team already runs — Linear, Notion, GitHub, internal docs — and Reverbic calls them while distilling, so specs land with your team's context attached.
Push work to where it lives
Meetings in, work out: summaries to Slack and Notion, recordings in from Zoom, code to GitHub, and everything else through MCP and your agents. More destinations are labeled honestly below — no logo-wall vaporware.
Connected today
On the way
Built for the security review, not just the demo
In production today
- AES-256 encryption at rest (Postgres + S3), TLS 1.3 in transit
- Org-scoped Postgres row-level security — cross-tenant reads blocked at the database layer
- No model training on your data — all AI via API
- Repo writes are Atmos-only, OAuth-scoped, and preview-only until you confirm
- Scoped MCP API keys with instant revoke
On the compliance roadmap
- SOC 2 Type II
- SSO / SAML for enterprise identity
- Zero-retention agreements with AI providers
- HIPAA support for regulated teams
Where this is going
The endgame is multi-agent: one planning call fans out into parallel agent tasks, and your agents sit in the meeting answering questions over MCP. Here's exactly what's shipped, what's being built, and what's still a plan — labeled honestly.
Shipped
- Atmos — live board → Claude-drafted PR, in every workspace
- Ask Reverbic — cited answers from any meeting
- MCP server — 5 tools, scoped keys, instant revoke
- rvb CLI — pure bash, distill from the terminal
- Slack, Zoom, Notion, GitHub integrations
- Desktop system-audio capture (macOS + Windows), iOS + Apple Watch
In progress
- Repo embeddings — grounding that retrieves by meaning, not just code search
- Agentic generation: Atmos requests extra files mid-diff instead of one-shot
- Real line-level diffs + per-file toggles in PR review
Planned
- Multi-agent fan-out: one planning call → parallel Claude Code tasks per action item
- Agents in the standup: your agent answers "what's blocking?" live over MCP
- Push action items to Asana, Jira, Monday, Linear
- REST API (today the CLI and MCP server are the developer surface)
Questions engineers actually ask
What models touch my audio and code?
OpenAI Whisper transcribes audio. GPT-4o writes summaries and distills specs, with a second verification pass that checks every requirement against the transcript. Atmos drafts code with Anthropic Claude, grounded in your repo. All via API — none of your data is used to train models.
Does Reverbic write to my repo?
Only Atmos does, only through GitHub OAuth you connect, and generation is preview-only: nothing is committed until you review the proposed files and confirm. The output is an ordinary pull request you can reject.
Is there an API?
The developer surface today is the CLI (curl install, pure bash) and the native MCP server at reverbic.ai/api/mcp with scoped keys. A REST API is on the roadmap — we'd rather tell you that than fake the docs.
What does the free plan actually include?
5 spec distills, 3 hours of transcription, and 10 meetings per month, for one user — plus Context Packs and all three export formats. No credit card. The Team plan removes the caps and adds the MCP server, Coach, and up to 20 members flat.
Where does my data live?
Postgres and S3, encrypted at rest with AES-256, TLS 1.3 in transit. Every table is row-level-security scoped to your org, so cross-tenant reads are blocked at the database layer — not in application code.
Stop retyping
your meetings.
The spec was said out loud an hour ago. Hand it to your agent.
curl -fsSL https://reverbic.ai/install.sh | bash