Orca Dev Stack
AdvancedRun Claude Code, Codex, and other coding agents side by side in Orca, each on its own branch and model.
Published 27 September 2026
About Orca Dev Stack
Orca is an open-source agent development environment that runs several coding agents in parallel, each in its own git worktree with its own branch, terminal, and file state. It doesn't wrap a model of its own: it hosts the agent CLIs you already use, signed in with your own subscriptions, so one agent can refactor the API layer while another writes tests, and nothing collides until you merge. Orca is free, MIT-licensed, and runs on macOS, Windows, and Linux.
In this stack agents and models are separate choices, and both allow more than one pick. Claude Code is the default agent, with Codex, OpenCode, GitHub Copilot CLI, Cursor CLI, and Grok Build alongside it, out of more than 25 that Orca supports. Each agent has a home model family, Claude for Claude Code and OpenAI's for Codex, but the model layer isn't welded to the agent: OpenCode runs almost any provider, Copilot CLI and Cursor CLI offer a model menu, and Codex can be pointed at other providers too. A common split is a premium model on the hardest task and a cheaper one like DeepSeek on bulk work.
Review happens inside Orca. Inline diff comments feed notes back to the agent that wrote the change, Design Mode sends a clicked UI element with its CSS and a screenshot into an agent's prompt, and pull requests, issues, and project boards from GitHub and Linear show up in the app. GitLab works as a plain git remote, with merge requests reviewed in GitLab itself.
Most people run Orca on the computer they work at, so the agents run there too and that machine has to stay on and awake while they work; the iOS and Android apps let you check progress and send the next instruction from your phone in the meantime. To keep agents going with the laptop shut, Orca can run headless on a Linux VPS or home server: the desktop app connects to it and shows the same worktrees, terminals, and diffs, while the agents and their compute stay on the server, and the phone apps pair with it directly or over Tailscale if you use it. Individual worktrees can also run on any machine Orca reaches over SSH. CI/CD is a common next step once several agents open pull requests at once.
Key Features
- ✓Several coding agents running in parallel, each isolated in its own git worktree and branch
- ✓Pick one or more agents: Claude Code, Codex, OpenCode, GitHub Copilot CLI, Cursor CLI, and Grok Build, with 25+ more supported
- ✓Pick one or more models, from Claude, OpenAI, and Grok to Gemini, DeepSeek, and Kimi through model-agnostic agents
- ✓Design Mode browser that sends UI elements, CSS, and screenshots straight into an agent's prompt
- ✓Inline diff comments that feed review notes back to the agent that wrote the change
- ✓Native GitHub and Linear panels for pull requests, issues, and project boards
- ✓Desktop apps for macOS, Windows, and Linux, plus iOS and Android apps for steering agents remotely
When to Use Orca Dev Stack
- →Splitting a large feature across several agents working on separate branches at once
- →Giving Claude Code and Codex the same task and merging whichever result is better
- →Routing hard reasoning work to a premium model and repetitive work to a cheaper open-weight one
- →Frontend work where pointing at a UI element is faster than describing it in a prompt
- →Keeping long-running agent sessions alive on a server and checking in from a phone
Pros
- Orca itself is free and MIT-licensed, with no per-seat fee and no proxy between you and the model providers
- No lock-in to one agent vendor: worktree isolation works the same way for every supported agent CLI
- Mixing models per task lets cheaper models absorb bulk work and lowers total token spend
- Parallel runs are safe to try, since nothing touches the main branch until you merge it
Cons
- Every agent needs its own subscription or API key, so costs add up across providers
- Reviewing several agents' output at once is more demanding than reviewing one
- Worktrees, multiple agents, and multiple models add concepts a single-agent setup never needs
- Using a model outside an agent's home family takes a model-agnostic agent or extra configuration
Coding Agent Options for Orca Dev Stack
Orca Dev Stack with Claude Code
Claude Code is the default agent here: Anthropic's CLI, strong at multi-file changes and a natural pick for the hardest task in a parallel run. It bills through a Claude plan from $20/month or per token via the API. It's built for Claude models; DeepSeek and GLM publish Anthropic-compatible endpoints it can point at, outside Anthropic's support.
Orca Dev Stack with Codex
OpenAI's agent CLI, included with every ChatGPT plan down to the free tier. Running Codex next to Claude Code on the same task is the most common way to compare two vendors and keep the better branch. It defaults to OpenAI's models but officially supports other providers through its config, including OpenRouter and local models.
Orca Dev Stack with OpenCode
Open-source and model-agnostic, OpenCode is what makes the model layer flexible: it runs Gemini, DeepSeek, Kimi, local models, or nearly any other provider with an API key, billed per token by that provider. It's the usual pick for the cheap, high-volume half of a parallel run.
Orca Dev Stack with GitHub Copilot CLI
GitHub's Copilot CLI, a good fit when a Copilot plan is already paid for: one subscription from $10/month covers a menu of Claude, GPT, and Gemini models, switched per session and metered in GitHub AI credits. It also ties naturally into the GitHub issues and pull requests Orca shows.
Orca Dev Stack with Cursor CLI
Cursor's agent running outside the editor, for developers already on a Cursor plan ($20/month and up). It brings Cursor's model menu, Grok by default with Claude, GPT, and Gemini available, into the parallel mix without another subscription.
Orca Dev Stack with Grok Build
xAI's coding CLI, which splits a task across its own sub-agents in separate worktrees. Add it for a third vendor's take on a hard task; it's built around xAI's Grok models and bills through an xAI plan or API key.
These are highlighted picks. To see all the tools, check the AI Coding Agents category.
LLM Options for Orca Dev Stack
Orca Dev Stack powered by Claude
Claude is the default model, the one Claude Code is built for, and a common choice for the task that needs the most careful reasoning. It's also selectable in OpenCode, Copilot CLI, and Cursor CLI, so one Claude subscription or key can back more than one agent in the same run.
Orca Dev Stack powered by OpenAI
OpenAI's GPT models are Codex's home family and are also reachable through OpenCode, Copilot CLI, and Cursor CLI. Picking them alongside Claude gives a second vendor's approach to the same problem, and a ChatGPT plan covers Codex usage without per-token billing.
Orca Dev Stack powered by Google Gemini
Gemini's long context window suits repository-wide questions and large refactors, and its Flash models are fast and cheap for routine work. Reach it through OpenCode, Copilot CLI, or Cursor CLI, or through Google's own Gemini CLI, which Orca also runs.
Orca Dev Stack powered by DeepSeek
An open-weight model priced far below the premium providers per token, which makes it the usual choice for the bulk half of a parallel run: test generation, mechanical refactors, and documentation across many files. It runs through OpenCode, or behind Claude Code via DeepSeek's Anthropic-compatible endpoint.
Orca Dev Stack powered by Kimi
Moonshot's open-weight Kimi models are another budget option tuned for agentic coding, with long context for large repositories. Use them through OpenCode or through Kimi's own CLI, which Orca supports directly.
Orca Dev Stack powered by Grok
xAI's Grok models, the family Grok Build is built around and Cursor's default. Add Grok when a Grok Build or Cursor CLI agent is in the mix, or reach it through OpenCode with an xAI API key for a fast option on routine tasks.
These are highlighted picks. To see all the tools, check the LLM category.
Version Control Options for Orca Dev Stack
Orca Dev Stack with GitHub
GitHub is the default here: Orca's pull request, issue, and project-board panels are built for it, so reviewing several agents' branches happens inside the app instead of in a browser tab.
Orca Dev Stack with GitLab
Worktrees work with any git remote, so GitLab is fine when source control has to be self-hosted. Merge requests are then reviewed in GitLab itself rather than in Orca's built-in panels.
Orca Dev Stack Add-ons
Each addition below extends this stack with a capability the base stack works fine without. None are required: include the ones your product actually needs when building this stack, and skip the rest.
Server Add-ons
Add a server when you want the agent to keep running somewhere other than your own laptop, reachable at any time and not tied to your machine staying on. Some agents also offer their own managed cloud sessions as an alternative to self-hosting; check the stack's own description for details.
Orca Dev Stack with Hetzner
Hetzner is the default here: a 4 GB server, Claude Code's minimum for one agent, is about €5.49/mo, and 8 GB or more suits several agents at once for little extra. Orca's headless mode runs on both its x86 and Arm servers, and data centers in Germany, Finland, the US, and Singapore let you pick one close to you for a responsive desktop connection.
Orca Dev Stack with DigitalOcean
DigitalOcean's 4 GB Droplet is $24/mo and 8 GB $48/mo, more per gigabyte than Hetzner, in exchange for more regions, a polished dashboard, and one-click snapshots and backups. Its 1-Click Codex CLI and OpenCode Droplets arrive with two of Orca's agents installed; the $4 plan's 512 MB is too small for any agent.
Orca Dev Stack with Hostinger
Hostinger's KVM 1 has 4 GB of RAM at $6.49/mo introductory, renewing at $11.99/mo, and KVM 2 doubles the memory for parallel runs. Its Claude Code VPS template preinstalls Orca's default agent, leaving only Orca's headless mode to add, and its Kodee assistant helps manage the server from a chat.
Orca Dev Stack with Raspberry Pi
A Raspberry Pi 5 with 4–16 GB of RAM is a one-time hardware cost instead of a monthly bill, and Orca's headless mode supports its ARM64 Linux. It suits a home setup reached over Tailscale; memory caps the fleet, with 8 GB fitting two or three agents and 16 GB a few more.
These are highlighted picks. To see all the tools, check the Hosting & Cloud category.
Remote Access Add-ons
Add remote access when you want to reach an agent running on another machine — a VPS or a home server — without exposing it to anyone but you.
Orca Dev Stack with Tailscale
Add Tailscale when Orca runs headless on a server: it's the pairing route Orca's own docs recommend, reaching the server from your laptop or phone without exposing it to the public internet. The personal tier is free.
Code Review Add-ons
Add code review when you want an AI reading every pull request before it merges: it comments on the diff so bugs, security issues, and inconsistencies surface before a human has to catch them.
Orca Dev Stack with CodeRabbit
Add CodeRabbit when several agents are opening pull requests faster than one person can read them: it reviews every GitHub or GitLab diff with a summary and line-level comments, with a free tier for open-source repos.
Orca Dev Stack with Greptile
Add Greptile when parallel agents produce more diffs than one person can review: it indexes the whole repository so its comments carry context beyond the changed files, and it can be self-hosted for full control.
Orca Dev Stack with Cursor Bugbot
Add Cursor Bugbot when a bug-focused reviewer is enough: it reviews GitHub or GitLab pull requests whatever wrote the code, though it bills through a Cursor account and plan rather than its own subscription.
CI/CD Add-ons
Add CI/CD when you want a dedicated pipeline for running tests, linting, or multi-stage builds before a deploy goes out. Many hosting platforms already redeploy automatically on every push on their own — a CI/CD tool adds the most value on top of that by gating the deploy on a passing test suite, and matters even more when the hosting choice does not auto-deploy at all, such as a self-hosted server.
Orca Dev Stack with GitHub Actions
Add GitHub Actions when several agents are opening pull requests at once and every branch should be tested automatically before you spend time reviewing it.
Orca Dev Stack with GitLab CI/CD
Add GitLab CI/CD for the same automatic testing when the repo lives on GitLab, with pipelines defined in a .gitlab-ci.yml file that can run on your own runners.
These are highlighted picks. To see all the tools, check the CI/CD Pipelines category.
Frequently Asked Questions about Orca Dev Stack
Can agents keep working when my laptop is closed?
Yes, when Orca runs on a server instead of your laptop. Orca's headless mode runs on an always-on Linux server, and the desktop app connects to it with the full interface, worktrees, terminals, and diffs, while agents and their compute stay on the server, so closing the laptop doesn't stop them. The iOS and Android apps pair with the same server, directly or over a private network like Tailscale, though a headless host doesn't send push notifications. Without a server, the phone apps steer agents on your desktop, which has to stay on. Orca has no hosted cloud of its own.
Which agents can run which models?
Most combinations work, with varying effort. OpenCode takes almost any provider's key, and Copilot CLI and Cursor CLI each offer a menu of Claude, GPT, and Gemini models through one subscription. Codex defaults to OpenAI but officially supports other providers through its configuration, local models included. Claude Code is built for Claude, and Anthropic supports only Claude models, though DeepSeek and GLM publish Anthropic-compatible endpoints some developers point it at. A common setup is Claude Code on Claude for the hardest task plus OpenCode on DeepSeek for bulk work.
How do parallel agents avoid overwriting each other's work?
Each agent works in its own git worktree on its own branch, so their file edits never touch each other or the main branch. Conflicts can still appear when you merge, exactly as they would between two human branches, so it helps to hand each agent a task in a different part of the codebase.
Do I need a separate subscription for every agent I add?
Yes. Orca bundles no model access and doesn't proxy anything; each agent signs in with its own account or API key. Claude Code uses a Claude plan, Codex uses a ChatGPT plan, and OpenCode bills per token with whichever provider you connect. Copilot CLI and Cursor CLI are the exceptions that reach several providers' models through one subscription, which can be the cheaper way to mix models if you already pay for one of them.
How is Orca different from Conductor or Herdr?
Conductor also gives each agent its own worktree, with a visual diff viewer and one-click pull requests, but it's a Mac-only app with four agents and a paid hosted cloud rather than self-hosting. Herdr is a terminal multiplexer that keeps agent sessions alive and reattachable, without managing worktrees or offering a GUI. Orca covers the most ground of the three: cross-platform desktop and mobile apps, 25+ supported agents, and a built-in browser for frontend work.
Stacks Related to Orca Dev Stack
Paseo Dev Stack
Developer
Run Claude Code, Codex, and other coding agents on your own machine with Paseo and steer them from any device.
Claude Code Dev Stack
DeveloperClaude Code in the terminal, desktop app, or web, paired with Claude and the add-ons to run it anywhere.
T3 Code Dev Stack
Developer
One open-source app for Claude Code, Codex, OpenCode, and other coding agents, on desktop and mobile.
Conductor Dev Stack
Developer
Run Claude Code, Codex, and Cursor's agent in parallel on your Mac with Conductor, each in its own workspace.
Scores
Popularity3/5
Orca is the most-starred tool among agentic dev environments, but running fleets of coding agents in parallel is a young practice that most developers outside AI-heavy teams haven't adopted yet.
Learning Curve4/5
Running agents in parallel adds git worktrees, merge coordination, and reviewing several streams of output on top of learning each agent itself, and choosing models per task adds another decision a single-agent setup never needs.
Flexibility5/5
Any of 25+ agent CLIs can run side by side, model-agnostic agents open the model layer to nearly any provider, and Orca runs locally, headless on a server, or steered from a phone.
Performance4/5
Parallel agents finish independent tasks in the time one would take for a single task, and the WebGL terminal stays responsive across many sessions. Several agents plus an embedded browser use far more memory than one terminal agent.
Portability5/5
Orca is MIT-licensed and runs on macOS, Windows, Linux, iOS, and Android, and every piece underneath (agent, model, git host) can be swapped without changing how the worktree workflow operates.
Tools in the Orca Dev Stack Stack
Development Tools
Add-ons (optional — add any, or none)
Server
Remote Access
Code Review
CI/CD
Orca Dev Stack Pricing
Orca itself is free and open source, so the bill comes from the agents. Claude Code starts at $20/month on Claude's Pro plan with no free tier, Codex has a limited free tier with Plus at $20/month, and both reach $200/month on their heaviest plans; OpenCode is free but bills per token with whichever provider you connect. Two or three agents in parallel usually means two or three subscriptions, the biggest cost here. A server for headless runs is about €5.49/month for 4 GB and more for several agents.
MIT-licensed desktop, mobile, and headless apps with no per-seat fee.
Claude Code from $20/mo (no free tier); Codex free tier, Plus $20/mo; both up to $200/mo; Copilot Pro $10/mo. One subscription per agent.
OpenCode and other model-agnostic agents bill per token with the provider you connect; open-weight models like DeepSeek and Kimi cost a fraction of premium ones.
Free tier covers individual devs; GitHub Team is $4/mo, GitLab Premium is $29/mo for more seats.
Tailscale's personal tier is free for up to 100 devices.
4 GB runs one agent, 8 GB or more suits several: Hetzner from €5.49, Hostinger $6.49 intro ($11.99 renewal), DigitalOcean $24–48; or a Raspberry Pi as a one-time cost.