feat: add Astrai as intelligent LLM router provider

[beee003]

Summary

• Adds Astrai as a 4th LLM provider mode (LLM_PROVIDER=astrai) alongside openai-raw, ai-sdk, and cloudflare-gateway
• Astrai is an AI inference router that automatically selects the optimal model/provider for each request based on cost, latency, and task complexity
• Instead of locking into a single provider, Astrai routes across OpenAI, Anthropic, Google, Groq, DeepInfra, and more — finding the cheapest equivalent that meets quality requirements
• OpenAI-compatible API (/v1/chat/completions), so it drops in cleanly alongside the existing providers

Why this is useful for MAHORAGA

MAHORAGA already supports 5+ LLM providers, but users still have to manually pick one. Astrai adds intelligent routing — set LLM_MODEL=auto and it picks the best model per request. For a trading agent that makes hundreds of LLM calls daily (research, classification, analysis), this can significantly reduce costs by routing simple sentiment classification to cheap models (Groq/Llama) while escalating complex analysis to frontier models (GPT-4o/Claude).

Three routing strategies:

Strategy	Description
balanced	Balance cost and quality (default)
cheapest	Minimize cost while maintaining quality
fastest	Minimize latency

Changes

File	Change
src/providers/llm/astrai.ts	New provider — OpenAI-compatible client for Astrai's /v1/chat/completions
src/providers/llm/factory.ts	Add "astrai" case to provider factory + isLLMConfigured
src/providers/llm/index.ts	Re-export AstraiProvider
src/env.d.ts	Add ASTRAI_API_KEY, ASTRAI_STRATEGY env vars
README.md	Document the new provider mode with config examples

Usage

npx wrangler secret put LLM_PROVIDER    # "astrai"
npx wrangler secret put LLM_MODEL       # "auto" or specific model
npx wrangler secret put ASTRAI_API_KEY   # sk-astrai-...
npx wrangler secret put ASTRAI_STRATEGY  # "balanced" (default), "cheapest", or "fastest"

Test plan

• Verify existing providers (openai-raw, ai-sdk, cloudflare-gateway) still work unchanged
• Test LLM_PROVIDER=astrai with a valid API key
• Test LLM_MODEL=auto for automatic model selection
• Test fallback when ASTRAI_API_KEY is not set (should return null gracefully)
• Run existing test suite (npm test)

🤖 Generated with Claude Code

[greptile-apps]

Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b29234e290

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Wire astrai through agent config validation

The new astrai branch in createLLMProvider is not reachable from the harness configuration flow: handleUpdateConfig validates updates with safeValidateAgentConfig (src/durable-objects/mahoraga-harness.ts:1193), but AgentConfigSchema still only allows "openai-raw" | "ai-sdk" | "cloudflare-gateway" (src/schemas/agent-config.ts:28). Because initializeLLM prefers state.config.llm_provider over env, users cannot persist astrai via the existing config API, so the new provider mode is effectively unavailable in that workflow.

Useful? React with 👍 / 👎.