feat: add avalanche stealth provider

[steebchen]

Add avalanche stealth provider with Vertex-style authentication and URL structure.

Includes gemini-3-pro-image-preview, gemini-3-pro-preview, gemini-3-flash-preview, and gemini-3.1-pro-preview models with 20% discount. Requires LLM_AVALANCHE_API_KEY, LLM_AVALANCHE_BASE_URL, and LLM_AVALANCHE_PROJECT environment variables. Supports optional LLM_AVALANCHE_REGION (defaults to "global").

All gateway response processing handles avalanche alongside google-ai-studio and google-vertex providers for consistent behavior.

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

Summary by CodeRabbit

• New Features
  • Added Avalanche as a new provider with endpoint, header, and streaming support.
  • Integrated Avalanche into routing, auto-fallback, logging, and token accounting so it behaves like existing Google-style providers.
  • Added Avalanche model entries with pricing and capability metadata for multiple Gemini variants.

[coderabbitai]

Walkthrough

Adds "avalanche" as a Google-compatible provider across the gateway: routing, request/response transformation, streaming, token/accounting, endpoint/header construction, model/provider definitions, and extraction tools are updated so Avalanche follows the same code paths as Obsidian/Google providers.

Changes

Cohort / File(s)	Summary
Chat Routing & Capabilities apps/gateway/src/chat/chat.ts	Treats avalanche like Obsidian/Google in routing, uptime gating, streaming handling, token forwarding, and log enrichment branches.
Content / Image / Reasoning Extraction apps/gateway/src/chat/tools/extract-content.ts, apps/gateway/src/chat/tools/extract-images.ts, apps/gateway/src/chat/tools/extract-reasoning.ts	Added avalanche to existing Obsidian switch cases so content, images, and reasoning follow the same extraction paths.
Token Usage & Tool Calls apps/gateway/src/chat/tools/extract-token-usage.ts, apps/gateway/src/chat/tools/extract-tool-calls.ts	Included avalanche in Google-style token-usage and tool-call branches (uses usageMetadata and Google adjustments).
Provider Response Parsing & Transformation apps/gateway/src/chat/tools/parse-provider-response.ts, apps/gateway/src/chat/tools/transform-response-to-openai.ts, apps/gateway/src/chat/tools/transform-streaming-to-openai.ts	Added avalanche as an alias in Google-style parsing and OpenAI transformation branches, including streaming transformations.
Costs & Logs apps/gateway/src/lib/costs.ts, apps/gateway/src/lib/logs.ts	Expanded Google-style provider set to include avalanche for token accounting and unified finish-reason mappings (STOP→COMPLETED, UNKNOWN finish handling).
Provider Endpoint, Headers & Request Prep packages/actions/src/get-provider-endpoint.ts, packages/actions/src/get-provider-headers.ts, packages/actions/src/prepare-request-body.ts	Added Avalanche endpoint construction (env vars: LLM_AVALANCHE_*) and header behavior; integrated avalanche into Google-compatible request body preparation and header logic.
Provider & Model Metadata packages/models/src/providers.ts, packages/models/src/models/google.ts	Added avalanche provider entry and numerous Gemini-related skip-test model entries for Avalanche variants (pricing/feature metadata).

Sequence Diagram(s)

mermaid
sequenceDiagram
participant Client
participant Gateway as Gateway (router/transform)
participant Actions as Actions (endpoint/headers)
participant Provider as Avalanche/Google-Backend
participant Logs as Logs/Costs

Client->>Gateway: Send chat request (provider=avalanche)
Gateway->>Gateway: Recognize provider as Google-style (transform body)
Gateway->>Actions: build endpoint & headers (LLM_AVALANCHE_*)
Actions->>Provider: Forward request (stream or generate content)
Provider-->>Gateway: Return streaming/data responses
Gateway->>Logs: Enrich logs, compute tokens/costs
Gateway-->>Client: Return transformed OpenAI-compatible response

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• feat: add google-vertex provider support #1070: Adds a Google-compatible provider across provider registration, endpoint/header, request/parse, and routing logic — closely mirrors adding avalanche.
• fix(gateway): fix Google thinking token counting #1631: Modifies Google-style token parsing and cost accounting branches that this PR extends to include avalanche.
• feat(gateway): map content filter types #1345: Adjusts unified finish-reason mapping in logs.ts; related because this PR adds avalanche to the same finish-reason mappings.

Suggested reviewers

• smakosh

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'feat: add avalanche stealth provider' clearly and concisely describes the main change: adding support for a new Avalanche provider to the system.
Docstring Coverage	✅ Passed	Docstring coverage is 92.86% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch add-avalanche-provider

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Copilot]

Pull request overview

This PR adds support for the Avalanche stealth provider, a Google Vertex-compatible LLM provider with authentication and URL structure following the Vertex AI pattern. The provider offers Gemini models (3-pro-preview, 3.1-pro-preview, 3-flash-preview, and 3-pro-image-preview) with a 20% discount compared to standard pricing.

Changes:

• Added Avalanche provider configuration requiring API key, base URL, project, and optional region environment variables
• Added four Gemini model variants for Avalanche provider with tiered pricing and 20% discount
• Integrated Avalanche into all Google-style response processing logic for consistent behavior with other Google-compatible providers

Reviewed changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
packages/models/src/providers.ts	Added Avalanche provider definition with Vertex-style environment configuration
packages/models/src/models/google.ts	Added Avalanche provider mappings for gemini-3-pro-preview, gemini-3.1-pro-preview, gemini-3-flash-preview, and gemini-3-pro-image-preview with pricing tiers and 20% discount
packages/actions/src/prepare-request-body.ts	Added Avalanche to Google provider case for request body transformation
packages/actions/src/get-provider-headers.ts	Added Avalanche to return empty headers (token passed via query parameter)
packages/actions/src/get-provider-endpoint.ts	Added Avalanche endpoint generation with Vertex-style URL structure including project/region/publisher path
apps/gateway/src/lib/logs.ts	Added Avalanche to Google provider checks for finish reason handling
apps/gateway/src/lib/costs.ts	Added Avalanche to Google provider group for token counting and cost calculation
apps/gateway/src/chat/tools/transform-streaming-to-openai.ts	Added Avalanche to Google provider case for streaming response transformation
apps/gateway/src/chat/tools/transform-response-to-openai.ts	Added Avalanche to Google provider case for non-streaming response transformation
apps/gateway/src/chat/tools/parse-provider-response.ts	Added Avalanche to Google provider case for response parsing
apps/gateway/src/chat/tools/extract-tool-calls.ts	Added Avalanche to Google provider case for tool call extraction
apps/gateway/src/chat/tools/extract-token-usage.ts	Added Avalanche to Google provider case for token usage extraction
apps/gateway/src/chat/tools/extract-reasoning.ts	Added Avalanche to Google provider case for reasoning content extraction
apps/gateway/src/chat/tools/extract-images.ts	Added Avalanche to Google provider case for image extraction
apps/gateway/src/chat/tools/extract-content.ts	Added Avalanche to Google provider case for content extraction
apps/gateway/src/chat/chat.ts	Added Avalanche to Google provider checks throughout chat handling for token passing, thought signature caching, web search counting, image counting, finish reason handling, and content filtering

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

apps/gateway/src/chat/tools/transform-streaming-to-openai.ts (1)

254-255: Optional: fix the hardcoded "google-ai-studio" in the extractImages call (line 431)

Adding obsidian and avalanche to this branch means line 431 now sends the wrong provider string to extractImages for two providers. It still works today because all four providers in this branch share the same code path in extractImages, but it's a latent correctness risk if those paths ever diverge.

♻️ Suggested fix

- delta.images = extractImages(data, "google-ai-studio");
+ delta.images = extractImages(data, usedProvider);

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@apps/gateway/src/chat/tools/transform-streaming-to-openai.ts` around lines
254 - 255, The switch branch that now handles case "obsidian" and case
"avalanche" is calling extractImages with a hardcoded provider string
"google-ai-studio"; update the call so it forwards the actual provider
identifier (the variable used in the switch) instead of the hardcoded literal.
Locate the cases for "obsidian"/"avalanche" and the extractImages(...)
invocation in transform-streaming-to-openai (the call site referencing
extractImages) and replace the static "google-ai-studio" argument with the
provider variable used in the surrounding function so each provider passes its
correct provider name to extractImages.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@packages/models/src/models/google.ts`:
- Around line 767-803: The Avalanche entry for modelName
"gemini-3-flash-preview" (providerId "avalanche") has incorrect numeric
constants and an unexpected pricingTiers block: change contextSize from 1050000
to 1048576 and maxOutput from 65500 to 65535 to match sibling entries
(google-ai-studio/google-vertex), and remove the pricingTiers array entirely
unless Avalanche intentionally uses tiered pricing—if it is intentional, keep
the block but confirm the tier values are correct and documented.

---

Nitpick comments:
In `@apps/gateway/src/chat/tools/transform-streaming-to-openai.ts`:
- Around line 254-255: The switch branch that now handles case "obsidian" and
case "avalanche" is calling extractImages with a hardcoded provider string
"google-ai-studio"; update the call so it forwards the actual provider
identifier (the variable used in the switch) instead of the hardcoded literal.
Locate the cases for "obsidian"/"avalanche" and the extractImages(...)
invocation in transform-streaming-to-openai (the call site referencing
extractImages) and replace the static "google-ai-studio" argument with the
provider variable used in the surrounding function so each provider passes its
correct provider name to extractImages.

[coderabbitai]

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Duplicate comments:
In `@packages/models/src/models/google.ts`:
- Around line 767-787: No change required: the model entry object for modelName
"gemini-3-flash-preview" already has corrected fields (contextSize, maxOutput,
pricing/flags) matching the google-ai-studio/vertex siblings; leave the object
(test/providerId/modelName/inputPrice/outputPrice/cachedInputPrice/requestPrice/discount/contextSize/maxOutput/reasoning/reasoningMaxTokens/streaming/vision/tools/webSearch/webSearchPrice/jsonOutput/jsonOutputSchema)
as-is and approve the change.