A control plane for coding agents
One workspace for every agent you run — Claude, Codex, Cursor, Grok, and OpenCode — with each session isolated, observable, and ready to ship.
Isolated sessions, real worktrees
Each session runs against its own checkout, with its own agent, transcript, and queue — so parallel agents never collide.
Three workspace modes
Remote git sessions clone or reuse cached repositories, local git sessions start fresh repos, and empty sessions give you chat-style work without a workspace.
Branch-aware by default
Start from an existing branch or cut a new one from any base. Switch, rebase, merge, and squash from the session UI.
Interactive terminals
Open named terminals inside the session worktree, watch agent-run commands, and jump in yourself when it helps.
Queued prompts
Type ahead while the agent is working. Queued prompts wait their turn and land in order — nothing gets lost.
Pick up where you left off
See which sessions finished while you were away, which worktrees have uncommitted changes, and how much of each agent's usage limits remain.
Sub-sessions and forks
Delegate a task to a child session that inherits the full conversation, or fork the whole thing to explore an alternative without disturbing the original.
Control that scales with trust
Keep a human in the loop where it matters, and automate the rest.
Approvals
Sensitive operations pause for your approval. Review the exact command or change, then allow or deny — from your desk or your phone.
Checkpoints
Sessions snapshot progress as they go. Roll back to any checkpoint when an agent goes down the wrong path.
Goals and limits
Define the exact outcome, constraints, and completion criteria per session. Agents stop when the goal is met — or the limit is hit.
From prompt to pull request
Aiva closes the loop: review the diff, commit with guidance, push, and open a PR — then watch CI turn green.
Pull request workflows
Guided commit and PR creation with configurable message prompts, reviewer assignment, and squash or merge strategies.
CI visibility
GitHub Actions and Azure Pipeline runs show up next to your sessions, with job and step detail when something fails.
Preview environments
Spin up isolated SideQuest preview environments from a branch to validate changes in a real deployment before merging.
Pluggable where it counts
Wire Aiva into the tools your team already runs.
Agent functions
Long-running structured tasks with JSON-schema validation and automatic repair, executed by Codex or Cursor function runners.
Skills and slash commands
Package reusable instructions as workspace skills and trigger common flows with slash commands in the composer.
MCP servers
Attach workspace-scoped MCP servers and let agents use your internal tools through a signed, auditable boundary.
See it on your own code
Point Aiva at a repository and let an agent take the first task — you keep the approvals.