OpenCode free tier cuts off service mid-task with no warning - expose remaining allowance / add countdown

Summary

The free tier of opencode/deepseek-v4-flash-free cuts off mid-task with no warning when usage is exhausted, and nothing exposes how close you are to the limit. I run the OpenCode Home Assistant add-on (not a developer — Home Assistant user). Yesterday the assistant was midway through validating a change to my automations when the service was withdrawn ("time up"). Until the service came back I could not access or repair my automations, and I had zero indication a cutoff was coming.

Environment

• opencode version: 1.18.13 (via the Home Assistant OpenCode add-on)
• Model: opencode/deepseek-v4-flash-free
• Free tier, no subscription
• Operating System: Home Assistant OS 18.2 (generic-x86-64), add-on container
• Terminal: n/a — accessed through the Home Assistant OpenCode add-on (embedded TUI)
• Install/channel: Home Assistant add-on
• Active plugins: none

Steps to reproduce

• Not reproducible on demand: the cutoff is usage-based and hit once after a session of normal use. The experience was: no indication of remaining allowance at any point, then a hard stop mid-task ("time up") with no warning, and no recovery until the service came back.

Problem

• Free-tier limits are usage-based and provider-side; they are not exposed anywhere a client can read. No env var, no status in the MCP surface, and opencode stats only shows local history — not remaining allowance.
• Because the limit is invisible, the cutoff feels random and can land in the middle of a task.
• For Home Assistant users the impact is concrete: a mid-task cutoff can leave configuration half-changed and control surfaces (e.g. all automations) unavailable until the service is back.
• Recovery depends on the assistant being back online — a fragile position to be in.

Request

Please expose remaining free-tier allowance so clients can show a countdown/warning before the cutoff:

• A quota/usage endpoint or CLI command (e.g. opencode quota) returning remaining requests/tokens and the reset window, and/or
• Remaining-allowance info on responses (like standard rate-limit headers), and/or
• A pre-cutoff notice sent before the limit is hit (e.g. "X requests left this window") instead of a hard stop mid-stream.

Even a rough "low quota" signal before the cutoff would let clients warn the user and finish at a safe point (for HA: save and validate the config) instead of dying mid-write.

Related

• #30229 — "Free usage exceeded" despite active Go subscription
• #7479 — free models offered "for a limited time"