Coding agents on your machine.
Mission control in the cloud.
Aiva runs the coding agents you already use — Claude, Codex, Cursor, Grok, OpenCode — on your own hardware, and gives you one place to start sessions, watch them work live, approve changes, and ship pull requests. From any browser, including your phone.
Everything an agent session needs
Start, steer, review, and ship. Every session keeps running whether or not your browser does.
Sessions and worktrees
Every session gets its own git worktree — remote repositories, local repos, or chat-style sessions. Agents never step on each other's changes.
Live transcripts
Watch the agent think, edit, and run commands as it happens. Queue follow-up prompts while a turn is still running — they land in order, nothing gets lost.
Approvals and checkpoints
Sensitive operations pause until you approve them — from your desk or your phone. Roll back to a checkpoint when an agent takes a wrong turn.
Agent functions
Turn prompts into schema-validated JSON you can wire into workflows, run by Codex or Cursor function runners — with automatic repair when output doesn't match.
Sub-sessions
Hand a task to a child session with the full conversation context, run several in parallel, and pull the results back when they're done.
Pull requests and CI
Commit, push, open pull requests, and watch CI — without leaving the session. GitHub and Azure DevOps supported.
Three steps to your first agent session
Your hardware does the work. The cloud keeps the score.
Connect your machine
Run the Aiva agent and bridge on your hardware. The bridge dials out to the cloud — no inbound ports, no VPN.
Start a session
Pick a repository, a branch, and an agent — Claude, Codex, Cursor, Grok, or OpenCode. Aiva spins up an isolated worktree and puts the agent to work.
Steer from anywhere
Watch the live transcript, approve actions, review the diff, and open a pull request — from your desk or your phone. Close the laptop; the session carries on.
Your repository stays on your machine
Repositories, worktrees, credentials, and the agent runtime live on your hardware. The cloud never mirrors your source tree — it keeps the session record, so the UI stays truthful across refreshes, reconnects, and outages.
- Outbound-only bridge connection — no inbound ports, no VPN
- Worktrees, git operations, and agent credentials stay local
- Sessions and transcripts survive anything short of deleting them
# No inbound firewall rules required
$ aiva-bridge \
https://cloud.example.dev \
<instance-api-key>
bridge connected → cloud hub
bridge forwarding → http://127.0.0.1:5087
cloud session started → worktree s/1
POST /agent/functions
{
"runner": "codex",
"schema": {
"type": "object",
"properties": {
"risk": { "type": "string" }
}
}
}
# Validated. Repaired. Returned as JSON.Structured work, not just chat
Agent functions turn prompts into reliable, schema-validated results you can build on. Combine them with skills, slash commands, MCP tools, and your issue tracker to automate the boring parts of shipping software.
- Codex and Cursor function runners
- Workspace skills, slash commands, and MCP servers
- Issue linking with your project-management provider
Plugs into your stack
Aiva meets your code, your tickets, and your tools where they already live.
Source control
GitHub and Azure DevOps — clone, branch, push, and open pull requests with credentials scoped to your workspace.
CI/CD
GitHub Actions and Azure Pipelines runs appear next to the session that caused them — down to the failing step.
Project management
Link sessions to Linear, Asana, GitHub Issues, or Azure Boards. Agents read the ticket, search the backlog, and comment on progress.
Swarm Surf browser
Give agents a real managed browser in the cloud — navigate, click, screenshot, and record — isolated per session.
SideQuest previews
Agents spin up isolated preview environments from a branch and validate their changes in a real deployment before merging.
MCP — bring your own
Attach workspace MCP servers and hand agents your internal tools, alongside Aiva's built-in cloud tools.
Put your agents to work
Connect a machine, start a session, and watch your first agent ship a pull request before lunch.