Same OpenCode runtime when you pick it. Continuum is the multi-provider cockpit around it, down to the Go quota gauge.
OpenCode is the open source coding agent from Anomaly, the team behind SST. It is MIT licensed, sits near 195,000 GitHub stars as of August 2026, and is no longer only a terminal app: it ships a TUI, a Tauri desktop app, VS Code and Zed extensions, a server plus SDK mode, LSP support, and 75+ providers through models.dev including local models. Continuum treats OpenCode as a first-class provider and hosts Claude Code, Codex, Cursor, Grok, and Gemini beside it, adding native Mac, web, iPhone, and Watch clients, worktree Code panes, multi-account isolation for Claude and Codex, live quota gauges (including OpenCode Go’s 5h, weekly, and monthly windows), and a local dollar ledger. OpenCode’s own auth.json is also where Continuum keeps connector keys for OpenRouter and Z.ai Coding, so the two share one secret store instead of inventing a second. New to the category? Start with AI coding agents.
Updated 2026-08-03 · Mac stable · Win/Linux desktop beta
You want OpenCode plus Claude and Codex peers in one sidebar, phone approve, worktree diff and PR panes, multi-account Max stacking, and a local cost ledger by repo, not the agent alone.
You want OpenCode’s own surfaces (TUI, desktop app, VS Code or Zed extension) and are happy without Continuum’s mobile clients, gauges, and multi-provider sidebar.
Continuum session on an OpenCode Go model for a cheap refactor; second session on Claude Max for the customer branch. One sidebar, two rails.
Approve the OpenCode plan from the iPhone during standup. The diff pane already shows the worktree when you get back to the desk.
Go’s weekly rail crosses 70% on the menu bar, so the next batch moves to an OpenRouter model without anyone discovering the cap the hard way.
Analytics show OpenCode and connector spend after the history parse; Claude gauges show weekly heat. One ledger answers “what did this repo cost.”
Open the OpenCode desktop app or the VS Code extension, pick a models.dev provider, and work in the surface OpenCode designed.
Stay inside OpenCode. Sessions, LSP, and /share cover the loop; there is no phone to approve from on this path.
Check the Go or Zen dashboard in a browser tab when limits feel close, then keep going.
Cost is your provider balance plus whatever the invoices say. No per-repo ledger unless you add one, and OpenCode alone is genuinely enough for a lot of people.
Install, connect, first session - steps you can run the same day.
Install Continuum. Keep your existing OpenCode install; Continuum uses it rather than shipping a second copy.
Confirm opencode auth list shows the providers you want. Continuum reads that same store.
Code tab → new session → pick an OpenCode model from the catalog, including connector vendors once configured.
Spawn a second session on Claude Max in the same sidebar so subscription-rail and BYOK work sit side by side.
On OpenCode Go? Continuum’s Usage tab reads the Go 5h, weekly, and monthly windows, so caps stop being a surprise.
Pair the iPhone over QR or Tailscale so plan-ready and interrupt work while OpenCode runs on the host.
Add an OpenRouter or Z.ai key in Continuum Settings, then confirm the model appears in the picker before spawning.
Friday: read cost by repo and decide which tickets belong on cheap open weights next week.
auth.json. Continuum Settings writes entries there rather than creating a second secret format.CLAUDE_CONFIG_DIR and CODEX_HOME) from OpenCode auth, so accounts cannot cross-bill.Same engine under your own keys. Continuum cells name the cockpit (panes, gauges, isolation, mobile); OpenCode cells name the surfaces and provider breadth it owns outright. Neither column re-scores model quality.
OpenCode is an agent runtime with real surfaces of its own. Continuum is the multi-device workbench that can host it. Continuum’s ModelCatalog merges OpenCode and connector models into the Code tab picker behind enable guards and live probes, so the picker reflects what is actually authenticated.
Continuum keeps Claude Code and Codex first-class, which matters when a subscription rail and a BYOK rail should run the same week. OpenCode stays OpenCode-shaped by design, and that focus is why its provider catalog is wider than Continuum’s named connector list.
Continuum Settings writes OpenCode auth entries for connector vendors, the same file opencode reads. Nothing forces you to keep an API key in a second ad-hoc format, and revoking in one place revokes in both.
OpenCode ships an official VS Code extension and a Zed integration. Continuum has no editor plugin: it is a standalone workbench. If your loop lives inside the editor, that is a real point for OpenCode and worth saying plainly.
Mobile approve, multi-account isolation, quota rails including OpenCode Go, and a local usage dashboard are Continuum products. If you only need the OpenCode TUI, stay on OpenCode and skip the workbench.
Stack recipe: keep OpenCode installed and let Continuum drive it as a provider. Continuum reads and writes the same auth.json OpenCode uses for connector vendors, and reads OpenCode Go’s usage endpoint for the gauges, so nothing is duplicated. Running opencode bare in a terminal still works exactly as before. Continuum does not fork or replace OpenCode; it hosts it beside other engines.
$0 for the app on Mac, web, iPhone, and Watch. Bring your own keys or the Claude and ChatGPT plans you already pay for, and Continuum takes no cut. Optional hosted inference: Plus $25/mo, Max 100 $100/mo, Max 200 $200/mo, Ultra $500/mo. OpenCode model usage still bills your own OpenCode balance or provider keys.
OpenCode is MIT licensed and free with your own keys. Two first-party options as of August 2026: Zen, pay as you go at zero markup starting from a $20 prepaid balance with auto top-up, which includes frontier models; and Go, a flat subscription at $5 for the first month then $10/mo covering roughly a dozen open-weight coding models with 5h, weekly, and monthly caps. Frontier models are not in the Go bundle.
Compare total cost of work, not sticker price: Go’s $10/mo only covers open weights, Zen bills per request, and a Claude or ChatGPT plan you already own costs nothing extra to drive through Continuum. See AI coding pricing compared.
OpenCode plans checked 3 August 2026 against their pricing page. Vendors reprice often - if you spot a stale figure, tell us and we will fix it. Continuum never marks up a session it runs under your own login.
Continuum ladder: Free app · Plus $25/mo · Max 100 · Max 200 · Ultra - full pricing. Also analytics, multi-account, devices.
Yes. OpenCode is MIT licensed and free to run with your own provider keys. The paid options are optional: Zen (pay as you go from a $20 balance, zero markup) and Go ($5 first month, then $10/mo for open-weight models). Continuum adds no charge on top of either.
As of August 2026, Go is $5 for the first month and $10/mo after that, with rolling 5h, weekly, and monthly usage caps across roughly a dozen open-weight coding models. Frontier models such as Claude and GPT are not part of the Go bundle; those come through Zen or your own keys.
Not really. Continuum hosts OpenCode as a provider and adds a multi-device workbench around it. If you only want the agent, use OpenCode. If you want mobile approve, worktree panes, quota gauges, and a spend ledger across several engines, use Continuum and keep OpenCode underneath.
No. Continuum drives your installed OpenCode and shares its auth.json for connector vendors. There is no fork, no vendored copy, and no second credential format to maintain.
Yes. Add the key in Continuum Settings; it is written to OpenCode’s auth store and the models appear in the Code tab picker once the provider probe returns. Z.ai Coding works the same way through its models.dev provider id.
Yes. Continuum polls OpenCode Go’s usage endpoint and renders the rolling 5h, weekly, and monthly windows as rails beside Claude and Codex, so one glance covers every rail you are burning.
Not through OpenCode itself, which has no first-party mobile client. Continuum’s iPhone and Watch apps pair to the host over QR or Tailscale and can approve plans, interrupt a turn, and read the diff while the OpenCode session runs on the machine.
In OpenCode’s own auth store, typically auth.json under the OpenCode data directory. Continuum writes entries there and reads them back; it never copies secrets into a Continuum-specific file for those connectors.
OpenCode’s database rows carry a provider id, but live streaming events do not always. Continuum attributes in-flight spend to OpenCode and re-attributes it on the next history parse. It is a documented nuance, not a hidden markup.
No. Claude Code and Codex work without it. OpenCode matters when you want that runtime specifically, or a connector vendor such as OpenRouter or Z.ai Coding that rides its auth store.
Same form factor, opposite defaults on trust and autonomy. And a strong case for running both.
Read guide → GuideThe JSONL files on your disk already record every token. Here is how to turn them into a number you can defend.
Read guide → GuideClaude Code pricing when the weekly cap fires: Pro vs Max vs API, two stacked subscriptions, or Get Plus.
Read guide →Free app. Your subscriptions. Optional hosted inference. Mac stable - Windows and Linux desktop are beta.
vendor-neutral · local-first · multi-device