Your team and coding agents, on the same task.
Product, QA, support and engineering bring work to one shared thread. An agent reads the code, writes a plan and asks when it's unsure. The team comments and revises until it's right, then the agent builds it and opens a pull request your engineers review as usual.
Built for everyone who has a say in the product.
QA finds the bug. PMs want the copy change. Support hears what's confusing. Customer success wants the promise kept. In Viberglass they all bring their work to the same place, the agent does it with them, and engineers step in where their judgement matters: the plan and the pull request.
QA & Support
Ship the bug reports and copy fixes you've already written three times this quarter. The agent reads your repro steps, asks what it can't work out, and hands the fix to engineering as a pull request.
Product & Design
Push the small roadmap items engineering keeps deferring. Copy tweaks, button labels, spacing adjustments, empty states. Describe the change, review the plan, land the PR.
Sales & Customer Success
Close the loop on customer feedback without waiting on a sprint. Put what the customer asked for in a task, answer the agent's questions, and follow it until it's merged.
If your team reviews PRs on GitHub and has more good ideas than developer hours, this is for you.
From a request to a merged pull request, together.
Anyone asks
Whoever knows what needs to change writes a task in the web app or straight from Slack, and brings in the people who should weigh in. No repo access, no dev environment.
The agent plans, and asks
The agent reads the code and writes a plan: what it found and what it would change. When something's unclear, it asks the right person in the thread and waits for the answer.
The team agrees, the agent builds
Reviewers comment on lines of the plan and ask for revisions. When it's right, the agent builds it and opens a pull request your engineers review like any other. Merge it and the task closes.
Everyone who cares about the change, in one place.
One thread. Everyone in it.
Messages, the agent's questions, the plan and every revision, comments, runs and pull requests, in order, for everyone on the task. @mention a designer, ask a reviewer, let support watch. Nobody has to piece together what happened from five tools.
Agree on the plan. Then build it.
Comment on any line of the plan, suggest an edit, ask the agent to revise. Older versions stay readable and comparable. When the plan is right, start the build. Large plans can be built in parts, with a pull request for each.
Always clear whose move it is.
Home shows what's waiting on you: a question from the agent, a review request, a mention, a failed run. Notifications reach you in Slack or by email, so an agent isn't left waiting on an answer nobody saw.
Bring your own agent. And your own model.
Claude Code, Codex, Antigravity, Qwen Code, Mistral Vibe, OpenCode, Kimi Code and Pi, side by side. Use provider keys or coding plans, or point OpenCode and Pi at any OpenAI- or Anthropic-compatible endpoint, including a model you host yourself. Each run gets its own container.
Your infrastructure. Every run on the record.
Self-host with Docker Compose, Kubernetes or AWS. Your code goes only to the model provider you choose, or nowhere if you host the model. Every run shows what the agent thought and did, the prompt it got, its tool calls and what it cost.
Slack as the front door.
Start a task with /viberator and get a Slack thread that stays in step with it. Replies become messages on the task, the agent's questions come with answer buttons, and Build it starts the build when the plan is right.
Work on tasks from your AI tools.
The MCP endpoint lets Claude, Cursor or any MCP client list spaces and tasks, create a task, ask the agent for a plan or a build, and review its work. One API token, no glue code.
Schedules for the recurring work.
Run agents on a schedule to update dependencies, apply security patches, or do any other recurring job.
Chrome extension. Capture bugs where they happen.
Take screenshots, select elements, record your screen and annotate, all from the browser. The extension captures console logs, network errors, viewport size and browser details, and files a task with all of it.
It plugs into the tools you're already using.
Create tasks from GitHub Issues or Shortcut, capture bugs from the browser, start and discuss tasks in Slack, or push events with custom webhooks. Pull requests go to GitHub.
Frequently asked questions
Free and open source. Help if you want it.
Every feature is in the open-source release. You run it, and you pay only for your own infrastructure and models.
Self-hosted
FreeAGPL-3.0
Run Viberglass on your own infrastructure.
Every feature, no limits on tasks, people or agents
All eight agent harnesses and any compatible model endpoint
Docker Compose, Kubernetes or AWS
Community support on GitHub
Talk to us
Let's see
For teams that want a hand.
Help with installation and rollout
Custom integrations
A commercial license, if the AGPL doesn't fit
Bring the whole team into the work.
Self-host it for free on Docker, Kubernetes or AWS. Your team stops waiting on developer time for the things anyone can describe, and your developers stop guessing what was meant.