v0.1.5

Ports the AgentRouter v2.5.0 recovery fixes to OpenCode.

• Monthly spend is now reported in USD; it was 500,000× too high.
• Mid-stream response.failed frames with no error payload get an upstream_cluster_crash error.
• Requests identify as opencode/agentrouter (...).
• Interrupted streams retry after 30s, up to 3 attempts.
• Duplicate header entries are collapsed.
• /agentrouter_check shows Beijing quota refresh countdowns.

v0.1.4

Hide the deprecated built-in agentrouter model list so only the plugin's live AgentRouter providers appear.

v0.1.3

Fix Missing required parameter: 'input[N].id' on Responses models (e.g. gpt-6.1-sol): send history statelessly (store: false) and preserve item_reference.id during sanitization.

v0.1.2

Changes

• Replace the /agentrouter subcommand interface with separate /agentrouter_status, /agentrouter_pricing, /agentrouter_sync, /agentrouter_check, /agentrouter_pacing, and /agentrouter_key commands.
• Show only GPT 6 Astra, GPT 6.1 Sol, GPT 5.6 Sol, and Claude Opus 5.5 in the model picker. Other model definitions and pricing discovery remain intact.
• Update documentation and CLI help.

Package version: 0.1.2.

Build passed; no tests or live provider requests were run.

v0.1.1

Changes

• Set explicit readable model names in the model picker.
• Preserve OpenCode and AI SDK request headers instead of overriding client identity.
• Retain OpenCode fetch middleware and shared request pacing.

Package version: 0.1.1.

Build passed; no tests or live provider requests were run.