Cap OpenRouter max_tokens, add budget-aware retry, and improve streaming UI

[Amer-alsayed]

Motivation

• Prevent OpenRouter failures caused by oversized max_tokens requests that exceed account credits and return errors like "This request requires more credits, or fewer max_tokens.".
• Make the server-side provider call resilient by retrying with a reduced, affordable token budget when the provider signals a credit/token error.
• Improve client-side streaming rendering to handle SSE chunk boundaries, reduce UI thrash, and enhance live code/math rendering during streaming.

Description

• In api/chat.js replace the hardcoded max_tokens: 42000 with a configurable cap via OPENROUTER_MAX_TOKENS (default 8192) and clamp request values using a safe minimum of 256 by default.
• Add parseBudgetFromError() and createOpenRouterCompletion() helpers and implement a budget-aware retry that detects token/credit errors and retries once with a reduced max_tokens (parsed affordable value or a smaller fallback, bounded by 4096).
• In assets/js/app.js fix SSE decoding by buffering partial chunks (sseBuffer), decode with streaming support, throttle render frame rate and math/highlight updates, add a streaming reveal class toggle for smoother incremental updates, and improve viewport anchoring during partial renders.
• In assets/css/style.css add the .assistant-message-text.streaming-reveal animation and supporting @keyframes to smooth visual reveal while streaming.
• Bump the CSS version query in index.html to force clients to pick up the updated stylesheet.

Testing

• Ran node --check api/chat.js to validate syntax of the modified server file and it succeeded.
• No further automated test suite was present or executed in this rollout; changes are focused on request-safety and runtime streaming behavior and should be smoke-tested in a running environment with an OpenRouter API key.

---

Codex Task

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
chaqgpt	Ready	Preview, Comment	Feb 17, 2026 0:10am

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.