Inside OpenCode 2.0: separate the interface, the model and the decision to act
A fresh Product Hunt launch with a documented client/server boundary, public interface code and useful lessons about planning before execution.
OpenCode 2.0 launched on Product Hunt on 13 October 2026. The launch-history page showed 240 points when checked on 14 October; the overview showed 272. These are captured discovery signals, not a final vote total or a claim of product quality. Daily rank displays differed between retrieved pages, so no final ranking is asserted.
A recent launch, with a visible public boundary
OpenCode 2.0 is a new launch of an existing project, whose first Product Hunt launch was in July 2025. Its MIT licence permits software reuse with the required notices. That licence does not make every connected model service, hosted feature or downstream dependency free. This study checks published documentation and code; FreeStack has not benchmarked the agent or tested provider performance.
What the maintainers actually document
The official server guide describes a terminal UI communicating with an HTTP server, which exposes an OpenAPI 3.1 specification and can support other clients. The current public app manifest lists Solid and Tailwind; the desktop manifest lists Electron and its build tooling. These facts describe public modules, not undisclosed hosted infrastructure. The model guide names AI SDK and Models.dev as the provider layer and documents support for local models.
Planning is a different action from editing
The documented Build agent enables development actions. Plan is restricted, but its default edits and shell commands ask for approval rather than being impossible. An analysis-only exercise therefore needs explicit permission choices, not faith in a mode label. FreeStack's editorial lesson is to name the outcome, affected files and acceptance checks before any tool is allowed to change the project.
Try the small learning version
Prepare a developer handover with Excalidraw and VS Code: sketch one task, save the editable diagram, write a short Markdown brief and list three observable acceptance checks. Share that package for review before buying an agent subscription. This is an original FreeStack exercise inspired by the planning boundary, not OpenCode's production stack or a claim that these two tools reproduce the product.
What would justify paying?
The handover itself needs no AI membership. An editorial upgrade trigger is a recurring, defined implementation job that a chosen model can handle within a budget you understand. OpenCode's local-model support is not a promise of equal performance or zero hardware cost. Hosted models, provider subscriptions and inference charges need their own checks; first decide which requirement remains unsolved by the free manual workflow.
Sources & further reading
Follow the source for current details. Product Hunt is used for discovery and launch history; technical claims use provider documentation or the project repository.
- Product Hunt: launch history (13 October 2026 launch; checked 14 October)
- Product Hunt: overview and captured launch points
- Maintainer server architecture documentation
- Current application package manifest
- Current desktop package manifest
- Official model and provider boundary
- Official agent modes and permission defaults
- Explicit allow, ask and deny controls
- Maintainer MIT licence