[!IMPORTANT] This is a community extension. It is not affiliated with, officially maintained by, or endorsed by Kenari.

[!IMPORTANT] This is a modified fork of OnesoftQwQ/opencode-go-copilot, adapted for Kenari. Refactor was assisted with AI tools and reviewed by me.

Integrate Kenari into GitHub Copilot Chat as a VS Code extension.

The model list is read from Kenari's public GET /v1/models endpoint, so new models appear without an extension update. Each model is routed through the protocol it declares (OpenAI-compatible Chat Completions, OpenAI Responses, or Anthropic Messages), and reasoning effort levels are resolved from the catalog's reasoning_options.

Usage

1. Set API Key: Ctrl+Shift+P → Kenari: Set Kenari API Key and paste a kn- key from the Kenari dashboard
2. Show Models: Click the settings icon in the model picker → Language Models panel → set your desired models to Visible
3. Select Model: In the Copilot Chat bottom model picker, choose a "Kenari" model
4. Start chatting

Status Bar: Plan Quota & Token Indicator

The status bar shows your Kenari plan quota for the tightest rolling window (e.g. Kenari wk 16%, or Kenari -- before the first fetch). Hover the status bar item to see:

• Plan quota: the active plan, weekly / monthly utilization with reset countdowns, the active coupon, and your account balance.
• Token usage: cumulative input/output token counts, plus the cumulative cache hit count and cache hit rate for models that return cache metrics in an OpenAI-compatible format.

Click the status bar item or run Kenari: Check Kenari Plan Quota to refresh immediately. The tooltip usage section and the background refresh interval are controlled by kenari.showUsageInTooltip (default true) and kenari.usageRefreshInterval (default 5 minutes). The advanced token indicator can be disabled with kenari.enableThirdPartyTokenIndicator (default true); the native Copilot token indicator always stays visible.

[!NOTE] Reading quota and balance needs an unrestricted account key. Shared and restricted keys are refused by Kenari with HTTP 403, and the extension then keeps showing the last known values (or Kenari --).

Git Commit Messages

Click the magic wand button in the Source Control (SCM) panel to auto-generate a commit message.

You can configure the model, language, number of recent commits to reference, and whether to attach context files. The button can be hidden with kenari.enableCommitGeneration (default true).

Model Temperature Presets

Quickly switch temperature presets via Ctrl+Shift+P → Kenari: Set Model Temperature Preset.

Built-in presets:

You can also configure kenari.temperature and kenari.top_p directly in settings.json (requires kenari.modelPreset set to "custom").

Extended Vision Understanding

This extension adds extended vision understanding capability to text-only models that do not natively support vision. When you send a message with an image to these models, they can call a vision-capable model to answer specific questions about the image, and then answer based on the description.

You can configure the vision model and whether to enable thinking when describing images via kenari.visionProxyModel (default qwen3-7-plus) and kenari.visionProxyThinking (default false).

Model Catalog & Free Models

The model list and all metadata (context length, output limit, vision, reasoning levels, tool calling) come from GET https://kenari.id/v1/models, which is public and needs no API key.

• The list is cached for kenari.modelListRefreshInterval (default 60000 ms); run Kenari: Update Kenari Model List (forced) to refresh it manually.
• :free models are shown with a "Free" marker in the picker. They cost nothing to call but have per-minute and daily limits set by your account tier.
• The last successful list is cached on disk, so the model picker still works when Kenari is temporarily unreachable.

Reasoning Effort

Models that support thinking expose a reasoning effort selector in the model picker:

• Models with declared effort levels (e.g. low / medium / high / max) show those levels; a "Disabled" option appears only when the model actually accepts an off switch.
• Models with a simple on/off toggle show Disabled / Thinking.
• Models that always think show only their available levels.

Requests are mapped to Kenari's documented controls: reasoning_effort / reasoning on Chat Completions, output_config.effort / thinking on Messages, and reasoning.effort on Responses.

Output Token Budget

Kenari reserves balance for the worst case of a request before it runs: the estimated input plus max_tokens at the output price. Several models advertise an output ceiling just below their whole context window (e.g. 943,718 on a 1M window), so requesting that capability on every call would reserve the price of a full-length answer each time — and Kenari would silently lower max_tokens when the balance cannot cover it, cutting the answer short with finish_reason: length.

The extension therefore caps the requested max_tokens at 32,768 per call. The model's real capability is still declared to VS Code, and the upstream limit still applies; this only bounds what a single request asks for. If you need a longer single response than that, raise MAX_REQUESTED_OUTPUT_TOKENS in src/catalogModels.ts.

Error Handling and Retries

Failures are handled by HTTP status and Kenari's code (or error.type on Messages), never by the message text:

• Retried: 429 with rate_limit_exceeded, free_quota_rpm, upstream_error, passthrough_busy; plus 500, 502, 503 and 504. Retry-After is honoured when present; otherwise exponential backoff with jitter is used.
• Not retried: any 402 (insufficient_balance, payg_limit_reached, key_spend_cap_reached), free_quota_daily, plan_limit_reached, and rejections that need a different request (bad request, context length, content policy, entitlement).
• Failures that arrive after a stream has started (where the HTTP status is already 200) surface as errors too, instead of silently truncating the answer.

Proxy / Gateway (Advanced)

Advanced users can route inference requests (chat and Git commit message generation) through a self-hosted proxy/gateway for observability, key routing or compression. Run Kenari: Set Proxy Base URL and acknowledge the compatibility notice; the proxy must be fully compatible with the official endpoint (same protocols, paths, model IDs and headers, and unmodified streaming responses). Quota and model list requests always use the official endpoint. Leave the value empty to disable.

Commands

All commands are available from the command palette (Ctrl+Shift+P → Kenari: ...):

Configuration

Available in settings.json:

{
	"kenari.commitLanguage": "auto",
	"kenari.commitModel": "deepseek-v4-flash",
	"kenari.commitMessagePrompt": "",
	"kenari.requestTimeout": 600000,
	"kenari.recentCommitsCount": 10,
	"kenari.commitIncludeCommitDiff": false,
	"kenari.commitAttachContextFiles": true
}

Build

npm install
npm run compile
npm test           # runs the test scripts
npm run build      # packages extension.vsix