OpenCode Dev Stack
AdvancedOpenCode in the terminal, desktop app, or browser, running open-weight models like DeepSeek, GLM, and Kimi.
Published 27 September 2026
About OpenCode Dev Stack
OpenCode is an open-source, MIT-licensed coding agent from the team behind SST. It does the same job as Claude Code or Codex: it reads the repository, plans a change, edits files, runs commands and tests, and iterates on failures. The difference is that it isn't tied to one vendor, and this stack pairs it with open-weight models: DeepSeek by default, with GLM, Kimi, MiniMax, Xiaomi MiMo, and Qwen alongside, the lineup OpenCode's own Go subscription is built around. Each can come from its lab's API or coding plan, from OpenCode Go or pay-per-token Zen, or through an aggregator, and several can be in use at once: the build agent that edits files and the read-only plan agent can each run their own model, and a session can switch models mid-task.
The same agent runs in the terminal, where it started, in a desktop app for macOS, Windows, and Linux, in IDE extensions, and in a web interface. All of them are clients of one local OpenCode server, so they share the same sessions and settings. LSP support gives the agent the code intelligence an editor has, and MCP servers, custom tools, and plugins extend what it can do.
Most developers run OpenCode on their own computer, which then has to stay on. Its web interface serves those sessions to a phone or tablet browser on the same network, and community-built iOS and Android apps connect to it as well. There are no vendor-hosted cloud sessions, but OpenCode's GitHub integration runs the agent on GitHub Actions when an issue or pull request mentions it and returns a branch or pull request with your machine off. For sessions that run around the clock, OpenCode can also run headless on a VPS or home server, reached over SSH or, if you use it, Tailscale; DigitalOcean can even provision it preinstalled.
Code lives on GitHub by default, or on GitLab when source control has to be self-hosted. A faster terminal emulator, a model aggregator for one key across providers, CI/CD, and an AI code reviewer are common additions; none of them change how OpenCode works.
Key Features
- ✓Open-source, MIT-licensed agent across the terminal, a desktop app, IDE extensions, and a web interface
- ✓Open-weight models from DeepSeek, Z.ai, Moonshot, MiniMax, Xiaomi, and Alibaba, one or several at once
- ✓Models from each lab's API or coding plan, OpenCode Go or Zen, or an aggregator
- ✓Build and plan agents, each able to run its own model
- ✓GitHub integration that runs the agent on GitHub Actions from an issue or pull request comment
- ✓A web interface and headless server for reaching sessions from a phone or another machine
- ✓LSP and MCP support, custom tools, and plugins
When to Use OpenCode Dev Stack
- →Running a capable coding agent on open-weight models for a fraction of a premium subscription
- →Using a cheap model for routine edits and a stronger one for hard problems in the same project
- →Keeping the choice of model independent of any single vendor's pricing or policy changes
- →Handing issues to OpenCode through GitHub comments and reviewing the pull requests later
- →Running OpenCode on a home server or VPS and checking on it from a phone browser
Pros
- The agent is free, and open-weight models cost a fraction of premium models per token
- No lock-in: switch models or providers per task, or serve them yourself
- OpenCode Go covers the whole open-weight lineup for $10 a month
- Existing ChatGPT or GitHub Copilot subscriptions work in the same agent
Cons
- No vendor-hosted cloud sessions or official mobile app; away-from-desk work runs on hardware you keep on
- Open-weight models differ in how reliably they follow tool calls, so results depend on the model picked
- Claude Pro and Max plans can't be used; Claude needs a pay-per-token API key
- Models from several labs mean several billing accounts unless Go or an aggregator unifies them
- Some of these labs' own APIs process data in China, which some teams' policies rule out
LLM Options for OpenCode Dev Stack
OpenCode Dev Stack powered by DeepSeek
DeepSeek is the default here: V4.1-Flash is among the cheapest capable coding models, about $0.15 per million input tokens off-peak on DeepSeek's own API, with a 1M-token context. OpenCode connects to DeepSeek directly, and the model is also part of OpenCode Go, Zen, and OpenRouter.
OpenCode Dev Stack powered by GLM
Z.ai's GLM models are the usual flat-rate pick: OpenCode signs in to the GLM Coding Plan, from $18 a month, as well as the pay-per-token API, where GLM-5.3-Flash costs $0.15 per million input tokens. GLM-5.3 is the larger flagship with a 1M-token context, and both are in OpenCode Go.
OpenCode Dev Stack powered by Kimi
Moonshot's Kimi models are built for agentic coding: K2.7-Code at $0.95 in and $4 out per million tokens for everyday work, and K3 at $3 and $15 for the hardest tasks. OpenCode connects to Moonshot's API directly, and Kimi is also available through OpenCode Go and OpenRouter.
OpenCode Dev Stack powered by MiniMax
MiniMax-M3 pairs a 1M-token context and multimodal input with low per-token prices, plus a Token Plan subscription from $22 a month that works across coding agents. OpenCode lists MiniMax as a built-in provider, and M3 is also part of OpenCode Go and Zen.
OpenCode Dev Stack powered by Xiaomi MiMo
Xiaomi's MIT-licensed MiMo-V2.6 models do well on agentic coding for very little: $0.435 in and $0.87 out per million tokens for Pro, $0.14 and $0.28 for Flash. In OpenCode they're easiest to reach through OpenCode Go or Zen, or OpenRouter.
OpenCode Dev Stack powered by Qwen
Alibaba's Qwen family runs from cheap Flash and Plus tiers to the Qwen3.8-Max flagship at $2 in and $6 out per million tokens, and most releases are Apache 2.0 open weights, the easiest of these to serve locally with Ollama. OpenCode reaches it through OpenCode Go or Zen, OpenRouter, or a local server.
These are highlighted picks. To see all the tools, check the LLM category.
Version Control Options for OpenCode Dev Stack
OpenCode Dev Stack with GitHub
GitHub is the default here: the largest pull-request ecosystem, and the home of OpenCode's own GitHub integration, which runs the agent in GitHub Actions when someone mentions it on an issue or pull request.
OpenCode Dev Stack with GitLab
GitLab bundles source control, merge requests, and CI/CD in one self-hostable platform, the pick when code has to stay on your own infrastructure. OpenCode only needs git for local work, so everyday sessions are the same on either host.
OpenCode 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.
OpenCode Dev Stack with Hetzner
Hetzner is the default here: its smallest shared plan, CX23, has 4 GB of RAM for about €5.49/mo, double the 2 GB DigitalOcean recommends for OpenCode, so language servers and test runs fit alongside it. A plain VPS billed by the hour up to a monthly cap.
OpenCode Dev Stack with DigitalOcean
DigitalOcean offers a 1-Click OpenCode Droplet with the agent preinstalled and wired to its own inference service for open models like DeepSeek and Qwen; it recommends 2 GB of RAM, $18/mo with 2 vCPUs. Its Managed Agents, in public preview, also run OpenCode in isolated microVMs billed per second.
OpenCode Dev Stack with Hostinger
Hostinger's KVM 1 has 4 GB of RAM at $6.49/mo introductory, renewing at $11.99/mo, well above what OpenCode needs since the models run at the provider. A budget pick on a plain Linux image; Hostinger's agent templates target other tools, not OpenCode.
OpenCode Dev Stack with Raspberry Pi
A Raspberry Pi 5 is a one-time hardware cost instead of a monthly bill, and OpenCode ships ARM64 Linux builds. With the models served remotely, a 4 GB board is enough; 8 GB leaves room for builds and tests. It can also serve the web interface to your phone on the home network.
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.
OpenCode Dev Stack with Tailscale
Add Tailscale alongside a server to reach it privately from a laptop or phone: SSH into OpenCode's terminal interface, or open its web interface in a phone browser, without exposing either to the public internet. The personal tier is free.
Session Persistence Add-ons
Add session persistence when the agent runs on a server or over SSH: a terminal multiplexer keeps the session working after the connection drops or the laptop closes, and you reattach from any machine to pick up where it left off.
OpenCode Dev Stack with tmux
Add tmux when OpenCode runs on a server or over SSH: its terminal interface otherwise ends with the connection, while a tmux session keeps it working and lets you reattach from any machine. It also keeps the headless server behind OpenCode's web interface running after you log out.
OpenCode Dev Stack with Zellij
Add Zellij for the same keep-running sessions with friendlier defaults: shortcuts stay on screen, floating panes suit a test watcher beside OpenCode, and layouts come back after a reboot, though the running agent doesn't. Its web client reaches the terminal from a browser.
Terminal Add-ons
Add a terminal when you want a faster, more configurable place to run the agent than your OS default — most agent CLIs live here all day.
OpenCode Dev Stack with Ghostty
A GPU-accelerated terminal with native macOS and Linux integration, and one of the terminals OpenCode's docs name, a solid default for keeping its terminal interface open all day.
OpenCode Dev Stack with Alacritty
A minimal, GPU-accelerated terminal that does little beyond drawing text fast, also named in OpenCode's docs; pick it when a multiplexer or window manager already handles tabs and splits.
OpenCode Dev Stack with WezTerm
A GPU-accelerated terminal with built-in tabs, splits, and multiplexing configured in Lua, recommended in OpenCode's docs and able to stand in for both the terminal and tmux on a local machine.
OpenCode Dev Stack with Kitty
A GPU-accelerated terminal with its own graphics protocol and a scripting layer called kittens, another of OpenCode's recommended terminals, for developers who want deep customization.
OpenCode Dev Stack with iTerm2
The long-standing free macOS terminal, with split panes, profiles, triggers, and tmux integration. The familiar pick for Mac users who want OpenCode in the terminal they already know.
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.
OpenCode Dev Stack with CodeRabbit
Add CodeRabbit when you want an AI reviewer on every pull request OpenCode opens: it installs as a GitHub or GitLab app, summarizes the diff, and leaves line-level comments before anyone merges, with a free tier for open-source repos.
OpenCode Dev Stack with Greptile
Add Greptile when review should read the whole repository rather than just the diff: it indexes the codebase so its GitHub and GitLab comments carry wider context, and it can be self-hosted alongside your own models.
OpenCode 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.
Model Aggregator Add-ons
Add a model aggregator when you want one API key and one bill for models from many providers, with automatic fallback when one of them is down, instead of setting up each provider separately.
OpenCode Dev Stack with OpenRouter
Add OpenRouter when you want one key and one credit balance for every model here and hundreds more. OpenCode lists it as a built-in provider, requests fall back to another host when one is down, and it charges provider prices plus 5.5% on credit purchases, with a few free models at 50 requests a day.
OpenCode Dev Stack with LiteLLM
Add LiteLLM when you'd rather run the gateway yourself: an MIT-licensed proxy on your own server that puts DeepSeek, GLM, Kimi, and local models behind one OpenAI-compatible endpoint, which OpenCode connects to as a custom provider, with per-key budgets and spend logs. Self-hosting it is free.
These are highlighted picks. To see all the tools, check the AI Model Aggregators category.
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.
OpenCode Dev Stack with GitHub Actions
Add GitHub Actions when the repo lives on GitHub and every push OpenCode makes should run the tests. It's also where OpenCode's GitHub integration runs, so the same account covers both.
OpenCode Dev Stack with GitLab CI/CD
Add GitLab CI/CD when the project is on GitLab and needs the same automated tests on every push, defined in a .gitlab-ci.yml file that can run entirely on your own runners.
These are highlighted picks. To see all the tools, check the CI/CD Pipelines category.
Containerization Add-ons
Add containerization when you want the app packaged the same way across local development, staging, and production, or need to deploy somewhere that isn't a managed serverless platform.
OpenCode Dev Stack with Docker
Add Docker when OpenCode should work in a reproducible, disposable environment, or when the project needs the same setup locally, on a server, and in CI. OpenCode itself is also published as a container image.
These are highlighted picks. To see all the tools, check the Containerization category.
Frequently Asked Questions about OpenCode Dev Stack
Can OpenCode keep working when my computer is off?
Not on OpenCode's own infrastructure: unlike Claude Code or Codex, it has no hosted cloud sessions. The closest thing is its GitHub integration, which runs the agent on a GitHub Actions runner when someone mentions it on an issue or pull request and comes back with a branch or pull request, so your machine can be off. For interactive work, OpenCode's web interface serves the same sessions to a phone or tablet browser, and community-built iOS and Android apps connect to it too, but the computer hosting the session has to stay on and online. Developers who want sessions running around the clock put OpenCode on a VPS or home server, which DigitalOcean offers with OpenCode preinstalled.
Can I use Claude or GPT models alongside the open-weight ones?
Yes, in the same agent. OpenCode signs in with a ChatGPT Plus or Pro subscription or a GitHub Copilot plan, so GPT models and Copilot's model menu come without a separate API key. Claude works through an Anthropic API key billed per token, but not through a Claude Pro or Max plan: Anthropic doesn't allow its subscriptions in third-party agents, and OpenCode stopped shipping the plugins that did it. A common mix keeps a cheap open-weight model on the plan agent and a proprietary one on the build agent for the hardest changes.
How does OpenCode compare to Claude Code and Codex?
All three are standalone coding agents with a terminal interface, a desktop app, and IDE extensions, so the difference is the model layer and who runs the infrastructure. Claude Code and Codex are built around their vendor's models, on a subscription that also pays for hosted cloud sessions and an official phone app. OpenCode is free and model-agnostic, so a cheap open-weight model can take routine work and a stronger one the hard tasks, with no usage limits beyond each provider's. What it gives up is the vendor cloud: away from the desk, it runs on hardware you control.
Is OpenCode Go enough, or should I pay a model lab directly?
It depends on how many models you use. Go covers the whole open-weight lineup for one price, with separate limits per model that reset every five hours, weekly, and monthly, so rotating between models stretches it further, and it can fall back to pay-per-token Zen credit when a limit runs out. A single lab's coding plan, such as Z.ai's GLM Coding Plan or MiniMax's Token Plan, costs more but gives larger quotas on that lab's models. Paying per token through an API or OpenRouter suits irregular use. The Pricing section has current figures.
Why does the same task go differently on different models?
Because OpenCode keeps the agent fixed and swaps only the model, the differences between models show up directly. Some follow tool calls and edit instructions more reliably, some handle a large repository better, and Flash-class models trade depth for speed and price. When a task stalls or the agent keeps repeating a failed edit, the model is the first thing to change: switching mid-session keeps the conversation, and a stronger model often finishes what a cheaper one started.
Scores
Popularity4/5
OpenCode is one of the most-starred open-source coding agents with a large community, and pairing it with open-weight models is a common setup among cost-conscious developers, though less mainstream than Claude Code or Codex.
Learning Curve3/5
A first task runs within minutes, but choosing models, a way to pay for them, and per-agent settings is more setup than an agent with one vendor's model built in.
Flexibility5/5
Seventy-plus providers, a separate model per agent, plugins, custom tools, MCP and LSP support, and four surfaces from terminal to browser, with no fixed model or plan.
Performance4/5
Results track the chosen model, and strong open-weight models handle multi-file work well, but there's no hosted cloud to run tasks off the local machine.
Portability5/5
MIT-licensed, runs on any machine, keeps sessions locally, and every model underneath can move to another provider or to your own hardware.
Tools in the OpenCode Dev Stack Stack
OpenCode Dev Stack Pricing
OpenCode is free and open source, so the bill is the models. OpenCode Go covers the whole open-weight lineup for $10/month with per-model limits; otherwise each lab's API bills per token, mostly well under $1 per million input tokens, or a lab's coding plan runs from about $18/month. OpenRouter adds 5.5% on credit purchases, and a self-hosted LiteLLM adds nothing. GitHub or GitLab are free for individuals, and an optional server runs about €5.49–18/month.
MIT-licensed agent across the terminal, desktop app, IDE extensions, and web interface; it can also sign in with an existing ChatGPT or GitHub Copilot subscription.
OpenCode Go $10/mo with per-model limits; OpenCode Zen and each lab's API per token (DeepSeek V4.1-Flash from $0.15/M input off-peak); GLM Coding Plan from $18/mo, MiniMax Token Plan from $22/mo.
Free tier covers individual devs; GitHub Team is $4/mo, GitLab Premium is $29/mo for more seats.
OpenRouter charges provider prices plus 5.5% on credit purchases; LiteLLM is free to self-host.
Tailscale's personal tier is free for up to 100 devices.
OpenCode is light because models run at the provider: DigitalOcean recommends 2 GB ($18/mo); Hetzner's smallest 4 GB plan is €5.49, Hostinger $6.49 intro ($11.99 renewal); or a Raspberry Pi as a one-time cost.