SpecWithCode vs OpenCode and cloud agents
SpecWithCode is a free desktop app for macOS, Windows and Linux with source available for inspection. It is one interface over the coding agents' own harnesses (Claude Code, Codex, GitHub Copilot, Cursor and more) on your own machines, signed in with the plans you already pay for. A coordinator splits work across worker threads, and the app tracks each subscription's limits where the provider reports them.
The facts about other products below come from their own sites, docs and GitHub repositories, checked on 16 October 2026. They change quickly, so follow the links for the current state. SpecWithCode is our product, so read this page with that in mind.
Harness or interface
A harness is the agent loop, tools, permissions and sign-in that a vendor builds around its model. OpenCode is a harness: it brings its own agent loop and talks to model APIs. SpecWithCode drives the vendors' own harnesses instead: it is an interface over those native harnesses with its own concept on top: a coordinator per project that plans and delegates to worker threads, usage limits across accounts, your own machines joined over Connect, and handoffs that keep the spec in git.
At a glance
| SpecWithCode | OpenCode | Cloud agents | |
|---|---|---|---|
| What runs the agent (the harness) | Each vendor's own harness; SpecWithCode is the interface and runs no agent loop of its own. Claude Code, Codex, OpenCode, Muse Code and Pi natively, Cursor through its official SDK (Cursor SDK), and Grok Build, Antigravity, GitHub Copilot CLI and Gemini CLI through ACP | Its own agent loop with build and plan agents, across 75+ model providers (providers, README) | The vendor's agent in the vendor's environment |
| Where your code runs | Your own computers | Your machine | A vendor-hosted container or VM that clones your repo (sources below) |
| Claude Pro, Max or Ultra plan | Through Claude Code's own sign-in on each machine | Its docs say Anthropic prohibits this and the plugin has not been bundled since v1.3.0 (providers) | Billed through each vendor's own plan |
| Other machines | Pair supported machines over SpecWithCode Connect (default), your network or SSH, with Tailscale as an optional direct route; move a project or continue a thread on another machine | opencode serve and opencode web listen on 127.0.0.1 by default and can bind a network address (server, web) | Runs in the vendor's cloud |
| Apps | macOS on Apple silicon, Windows 10/11 (x64, arm64), Linux (x64, arm64); Mac signed/notarized, Windows/Linux unsigned | Terminal, desktop for macOS, Windows and Linux, IDE extension (download) | Web, IDE and GitHub |
| Price and license | Free for every developer and team; no seat, agent or hour fees; Connect relay included for 3 machines (pricing); source available, no new project-wide license grant | Open source (MIT) (license) | Included in or billed by each vendor's plan |
Cloud agents here means:
- Codex cloud: "Codex creates a container and checks out your repo."
- Claude Code on the web: each session runs in an Anthropic-managed VM that clones your repository, unless your organization routes sessions to a self-hosted environment.
- GitHub Copilot cloud agent: an ephemeral environment powered by GitHub Actions.
- Cursor cloud agents: isolated cloud VMs that clone your repo from GitHub, GitLab, Azure DevOps or Bitbucket.
- Google Jules: "Jules runs in a virtual machine where it clones your code."
Why a flat plan suits heavy agent work
Claude Max costs $100 and Claude Ultra $200 a month, and both include Claude Code (Anthropic). ChatGPT Pro costs $100, $200 or $500 a month and includes Codex (OpenAI). Both have weekly limits; verified Claude Ultra accounts have no five-hour session limit.
API billing charges for every input and output token (Anthropic API pricing, OpenAI API pricing). A coding agent resends its growing context on every turn, so a working day of several agents produces far more tokens than a chat. For that kind of sustained use, a flat plan can cost less than the same work at API rates. For light use, API billing can be cheaper. Compare against your own bills with the calculator on the SpecWithCode page; we do not promise a saving.
SpecWithCode helps you use those plans fully:
- it shows session and weekly windows for every connected account;
- it can move a thread to another signed-in provider before a limit, and continue it on another machine;
- it reads Codex accounts and their banked resets from a CLIProxyAPI hub (for usage reporting; chat does not route through it). It never reads Claude accounts from a hub.
Anthropic does not let third-party apps offer Claude.ai login or route requests through Pro, Max or Ultra credentials on behalf of users, or collect and relay Claude.ai sign-in tokens; signing in to Claude Code itself is fine (Anthropic's terms). SpecWithCode runs the Claude Code CLI installed on your machine, which keeps its own sign-in; SpecWithCode never reads that sign-in.
Private by default
SpecWithCode needs no account and has no cloud service that stores your code or threads. Threads, plans and settings stay in a local database under ~/.specwithcode, and telemetry is off unless you turn it on. Prompts and relevant context still go to the model providers you choose; local execution does not mean offline AI. The optional SpecWithCode Connect relay for remote access stays off until you sign in to it.
Specs and increments
SpecWithCode works without any CLI. If the project uses SpecWeave, you can link a thread to an increment; its plan card then shows the spec and task progress read from spec.md and ledger.jsonl. The link is read-only: the app never rewrites your spec or ledger. The app does not create increments for each thread on its own. When the project has one, a thread can work from the increment's tasks and acceptance criteria rather than only from the chat prompt; see Tasks and acceptance criteria.
Pick OpenCode
when you want one open agent loop across many API model providers. SpecWithCode can also use OpenCode as one of its harnesses.
Pick a cloud agent
when work should keep running with your laptop closed and your repository already lives on GitHub.
Pick SpecWithCode
when you or your team want one coordinator to split work across several agents on your own computers, watch every subscription's limits in one place, let agents use your Chrome through the SpecWithCode Browser extension, and keep your code off anyone else's servers.
Download SpecWithCode or compare it with BridgeMind.