v0.5.3 - Context Bar Token Reporting Fix

v0.5.3 - Context Bar Token Reporting Fix

Fixed

• Token counting double-counting bug: Fixed provideTokenCount returning inflated token counts for message objects. The Copilot chat participant's tokenizer adds its own BaseTokensPerMessage(3) per message on top of what countTokens returns, but Aether was also adding a 3-token role overhead — causing each message to be counted with 6 tokens of overhead instead of 3. countLanguageModelMessageTokens now returns raw content tokens only.

• Context bar not updating for Aether models: The Copilot chat participant reads usage data from a LanguageModelDataPart with MIME type 'usage' in the response stream, but Aether never emitted this data part. Added emitUsageDataPart() which reads from TokenTelemetryTracker after each request and emits the usage JSON as a LanguageModelDataPart.json(...) with MIME 'usage', enabling the Copilot chat participant to report token usage to the ChatContextUsageWidget.

Added

• Token Usage Status Bar: Connected TokenUsageStatusBar to TokenTelemetryTracker so the context window usage indicator in the status bar automatically updates after each successful request. Previously the status bar existed but was never registered or wired to the telemetry tracker.

• Agent Workspace Detection: Added workspace.isAgentSessionsWorkspace proposed API type declaration and detection in extension activation. Logs whether the extension is running in an agent sessions workspace.

How Context Bar Token Reporting Works

The Copilot chat participant reads usage from a LanguageModelDataPart with MIME type 'usage' in the response stream. Aether now emits this data part after each request:

{
  "prompt_tokens": 150,
  "completion_tokens": 75,
  "total_tokens": 225
}

This enables the Copilot chat participant to report promptTokens/completionTokens to the ChatContextUsageWidget, filling the context bar correctly.

Full Changelog: v0.5.2...v0.5.3

Release v0.5.2

Fix API key validation issue for Compatible/Custom providers utilizing customHeaders or local endpoints.

v0.5.1

Fixed

• OpenGateway API key configuration: Fixed bug where the API key input field was not shown in the Aether Settings page for OpenGateway. Root cause: \�ccountManager.ts\ had \supportsApiKey: false\ for OpenGateway despite the provider requiring an API key. Also added missing \�piKeyTemplate\ to \opengateway.json\ to enable the API key input prompt.

v0.4.8 - VS Code LM API v5 Alignment

What's New

VS Code Language Model API v5 Alignment

• Complete type definitions for VS Code LM API v5
• Updated all provider implementations to use new API signatures
• Models now properly appear in VS Code's language model selection list

MCP Bridge Fixes

• Added CancellationToken parameter to tool invoke handlers
• Improved error handling for tool invocation

Provider Updates

• Added isUserSelectable: true and category to all model info objects
• Fixed model visibility in VS Code model picker

Changes

• 8 files changed, 479 insertions(+), 90 deletions(-)
• Updated vscode.proposed.chatProvider.d.ts with complete types
• Fixed GenericModelProvider, SeraphynProvider, MoonshotProvider, ZhipuProvider, and more

Thanks

Thanks to all contributors and testers!

Aether AI v0.4.7

DeepSeek V4 Thinking Content Auto-Detection

Fixed

• DeepSeek V4 thinking tokens: Fixed DeepSeek V4 models (, ) failing on multi-turn tool calls due to missing (OpenAI path) or (Anthropic path) in replayed assistant messages. Implemented automatic detection for DeepSeek V4 models, ensuring API compliance without manual configuration. Addresses both SDK paths and makes DeepSeek models work out-of-the-box.

Added

• DeepSeek auto-detection: Added automatic for DeepSeek V4 models in both static and dynamic providers
• ModelOverride includeThinking support: Extended interface to support configuration via

Aether AI v0.4.5

Aether AI v0.4.5

Added

• DeepSeek V4 models: Added deepseek-v4-flash and deepseek-v4-pro models to the DeepSeek provider, both with 1M context window support (936K max input / 64K max output).
• Llama-4 models: Added Llama-4 model support with 1M context window (936K max input / 64K max output) and image input capability.

Fixed

• Fixed Compatible Provider (aether.compatible) not registering with VS Code due to missing vendor declaration in package.json languageModelChatProviders. Custom models configured via Aether Compatible Provider Settings now appear in the Copilot Chat model dropdown.
• Fixed MCP bridge tool invocation error caused by accessing undefined vscode.CancellationToken.None, replacing with undefined to resolve error when invoking tools like read_file.

Changed

• Updated Anthropic thinking block handling: removed fake placeholder thinking blocks, enforced thinking blocks to appear first in assistant messages per Anthropic API requirements

Download from VS Code Marketplace: https://marketplace.visualstudio.com/items?itemName=OEvortex.aether-ai

v0.4.4

Changelog

All notable changes to this project will be documented in this file.

[0.4.4] - 2026-04-27

Added

• DeepSeek V4 context window: DeepSeek V4 models now support 1M context window with 64K max output / 936K max input. Added new detector isDeepSeekV4Model() in globalContextLengthManager.ts.
• GLM-5/5.1 context window: Updated GLM-5 and GLM-5.1 models to 204,800 total context with 64K output (same as Gemini3) / 140K input.
• Kimi K2.5/K2.6 context window: Updated to 262,144 context (256K binary) with 32K output / 229,376 input.
• MiMo v2.5 context window: Added new detector isMiMoV25Model() for MiMo v2.5 at 262,144 context / 32K output.
• MiMo v2.5 Pro context window: Added new detector isMiMoV25ProModel() for MiMo v2.5 Pro at 1M context / 64K output.
• OpenCode Qwen Plus models: Added new detector isOpenCodeQwenPlusModel() for OpenCode Qwen3.5 Plus / Qwen3.6 Plus at 262,144 context / 32K output.
• HY3 model support: Added isHy3Model() detector and configured 256K context (224K input / 32K output).
• OpenCode Zen Go provider: Added new opencodego provider with dynamic model fetching from models.dev API.

Changed

• Codex client headers: Made Codex Version header and User-Agent configurable via environment variables.
• Codex models: Updated to current models. Removed deprecated gpt-5.2* and gpt-5.1-codex-mini. Added gpt-5.5, gpt-5.4 (with medium/high/xhigh reasoning), gpt-5.4-mini (with low/medium reasoning).

Fixed

• MCP Bridge checkbox selection: Fixed issue where tool checkboxes in the MCP Bridge TreeView were not responding to clicks. Removed manageCheckboxStateManually flag to let VS Code handle checkbox state automatically, and removed unused event emitter from TreeDataProvider.

v0.4.3

Added

• MCP Bridge (VS Code LM Tools → MCP): Added MCP Bridge feature that exposes VS Code Language Model tools to external MCP clients (Claude Code, Claude Desktop, Cursor, Windsurf). Includes:
  • HTTP MCP Server using @modelcontextprotocol/sdk with StreamableHTTP transport
  • Dynamic vscode.lm.tools registry polling with configurable interval
  • Activity Bar TreeView with extension-grouped tools and checkboxes
  • Tool visibility management (per-tool and per-extension toggle)
  • Rich tooltips with metadata, input schemas, and invocation counts
  • Colored icons (green=exposed, gray=hidden, lock=disabled)
  • Status bar integration showing server status and tool count
  • Auto-start server option via aether.mcpBridge.http.enabled setting
  • Commands: Start/Stop/Restart Server, Refresh Tools, Copy MCP Config, Show Log
  • Context menu support for toggling tools and groups
  • Configuration: port (default 59684), filters, polling interval
• Baidu Qianfan provider: Added Baidu Qianfan provider (baidu) with OpenAI-compatible endpoint at https://qianfan.baidubce.com/v2, bundled config src/providers/config/baidu.json (includes ERNIE 4.0 Turbo 8K model), and automatic model discovery via /models.
• StreamLake provider: Added StreamLake provider (streamlake) with OpenAI-compatible endpoint at https://wanqing.streamlakeapi.com/api/gateway/v1/endpoints, bundled config src/providers/config/streamlake.json (includes GPT-4o model), and automatic model discovery via /models.
• Tencent Hunyuan provider: Added Tencent Hunyuan provider (tencent) with OpenAI-compatible endpoint at https://api.hunyuan.cloud.tencent.com/v1, bundled config src/providers/config/tencent.json (includes Hunyuan TurboS Latest model), and automatic model discovery via /models.
• Volcengine provider: Added Volcengine provider (volcengine) with OpenAI-compatible endpoint at https://ark.cn-beijing.volces.com/api/v3, bundled config src/providers/config/volcengine.json (includes example endpoint model), and automatic model discovery via /models.
• Xiaomi Mimo provider: Added Xiaomi Mimo provider (xiaomimimo) with OpenAI-compatible endpoint at https://api.xiaomimimo.com/v1, bundled config src/providers/config/xiaomimimo.json (includes Mimo V2 model), and automatic model discovery via /models.

Build

• Ran npm run sync-providers to register new providers and regenerate all provider artifacts.
• Created bundled provider configs with initial models to enable VS Code registration: src/providers/config/baidu.json, src/providers/config/streamlake.json, src/providers/config/tencent.json, src/providers/config/volcengine.json, src/providers/config/xiaomimimo.json.

0.4.2

What's Changed

• feat: add MiniMax-M2.7 by @the-byte-bender in #24

New Contributors

• @the-byte-bender made their first contribution in #24

Full Changelog: v0.4.1...0.4.2

0.4.1

[0.4.1] - 2026-04-20

Added

• Dialagram provider: Added Dialagram provider (dialagram) with OpenAI-compatible router at https://www.dialagram.me/router/v1, Anthropic/Claude router at https://www.dialagram.me/router/claude, and model listing endpoint /model.
• Routing.run provider: Added Routing.run provider (routingrun) with OpenAI-compatible router at https://api.routing.run/v1 and automatic model discovery via /models.
• Crof AI provider: Added Crof AI provider (crof) with OpenAI-compatible endpoint at https://crof.ai/v1 and automatic model discovery via /models.
• Requesty provider: Added Requesty provider (requesty) with OpenAI-compatible endpoint at https://router.requesty.ai/v1 and automatic model discovery via /models.
• AIML API provider: Added AIML API provider (aimlapi) with OpenAI-compatible endpoint at https://api.aimlapi.com/v1 and automatic model discovery via /models.
• Ofox provider: Added Ofox provider (ofox) with OpenAI-compatible endpoint at https://api.ofox.ai/v1 and automatic model discovery via /models.
• Helicone provider: Added Helicone AI Gateway provider (helicone) with OpenAI-compatible endpoint at https://ai-gateway.helicone.ai and automatic model discovery via /models.
• Portkey provider: Added Portkey provider (portkey) exposing OpenAI-compatible endpoint at https://api.portkey.ai/v1, Responses endpoint at https://api.portkey.ai/v1 (oai-response), and Anthropic-compatible endpoint at https://api.portkey.ai. Model discovery available via /models.
• EdenAI provider: Added EdenAI provider (edenai) with OpenAI-compatible endpoint at https://api.edenai.run/v3/llm and automatic model discovery via /models.

Fixed

• Removed Aether CLI and AVA Supernova provider artifacts; cleaned up build/test environment and removed stale CLI references.
• Fixed duplicate openai key in src/accounts/accountUI.ts which caused a build-time duplicate-object-key warning.

What's Changed

• Align CLI to @aetherAI scope and clean MCP/extension handling by @codex in #18

New Contributors

• @codex made their first contribution in #18

Full Changelog: v0.3.9...v0.4.1