◇
Shared remote environment
A VPS or container becomes the team's persistent workspace: terminals, live previews, streamed browsers, and a unified timeline where code, feedback, and experiments converge.
Collaborative development for the AI era
From chat to shipping, keep humans and AI agents in the same context.
Turn a VM or Docker container into a shared dev station with shared shells, live previews, and chat-native workflows. Self-hosted by default.
Discuss in chat. Execute in shared context.
Shipping in a single thread
Same moment, one source of truth
Without bunny
With bunny
Governed gateway, shared context
Without bunny
With bunny
Earlier proposals
Current version
No shared context
Shared context
Proposal: 2 devs and silos
Broken context
Relevant context
The shift
From local repos to shared, executable context
Instead of each developer working locally and synchronizing through GitHub, teams collaborate directly around a common remote environment, with chat, roles, and permissions intact.
◎
Chat-native workflows
Work from a channel with the same context, the same coding agent, and distinct rights for every contributor, whether engineers, designers, operators, or non-technical members.
⟡
Open & self-hosted
Install the tools you already use (shell, CLIs, scripts, Codex, Claude) through prompts and shared workflows in an environment your team controls.
Beyond code
Versioning that captures how software is actually made
The Git commit is often the only durable artifact explaining a project's evolution. But in modern workflows, meaning lives elsewhere: chat threads, agent prompts, collective decisions, tests run, errors encountered, and trade-offs accepted.
bunny adds a semantic layer on top of code versioning, connecting each important change to the discussions, goals, and interactions that led to it.
- Chat discussions
- AI agent prompts
- Decisions & trade-offs
- Tests & regressions
- Code changes
One collaboration graph, not just a commit log.
GraphRAG
Queryable context for every change
Enriched metadata is ingested into a GraphRAG, enabling questions that Git alone cannot answer.
Why was this commit made?
Which discussion led to this implementation?
What objective did this change serve?
Which agents or contributors influenced this decision?
Parallel validation
CI feedback at the speed of vibe coding
Current workflows are poorly suited to AI-assisted development. Teams iterate faster, rely more on agents, yet CI arrives too late, takes too long, and sometimes gets bypassed entirely.
bunny integrates a validation agent that works in parallel with development: continuously testing changes, detecting regressions, analyzing errors, and suggesting fixes directly in chat.
Create
Humans & coding agents ship changes
Validate
Validation agent tests in parallel
Discuss
Fixes & trade-offs in chat
Traditional CI
Slow, disconnected, often too late
Parallel editing
True collaboration without overwrites or conflicts
On a shared VM, many contributors work at once: engineers in their shells, multiple AI agents in parallel. A single checkout would mean constant collisions, with one session stomping another's files mid-edit.
bunny provisions a git worktree for every agent and every user shell. Each session gets its own working directory and branch on the same repository, locally on the machine for isolated edits, safe parallelism, and merge when ready.
- ✓One worktree per agent and per shell session
- ✓Shared object store with no duplicate clones
- ✓Parallel branches, integrated through your usual Git flow
One repo. Many trees. Everyone ships in parallel.
Governance
Authorization on every action, for everyone
bunny sits between your team and every connected tool with a verification and enforcement layer. Before any teammate or AI agent runs a command, opens a PR, or touches an integration, bunny checks authorizations against your policies and RBAC rules.
The same governance applies whether the request comes from a human in chat or an autonomous agent, with no backdoors or shadow access. Policies travel with context across GitHub, shells, browsers, and every MCP-connected service.
- ✓Role-based access for humans and agents
- ✓Policy enforcement on every connected tool
- ✓Centralized audit trail in chat context
One gate. Every tool. Humans and agents alike.
Collaborate around context, not just a repository
Software development as a living process where humans, AI agents, environments, discussions, tests, and decisions stay connected in one flow.
Get started on GitHub