Manifest Weekly #02

This was a big week. We added two new subscription providers, shipped a smarter model registry and fixed a pile of bugs that were affecting real users.

What we shipped this week

GitHub Copilot subscription provider You can now connect your GitHub Copilot subscription to Manifest. Authenticate via GitHub’s device login flow, and Manifest discovers all the models your Copilot plan gives you access to: Claude, GPT, Gemini, and more. All at $0 since your subscription already covers them. This means your OpenClaw agents can route through Copilot models without any additional API spend.

MiniMax subscription provider MiniMax is now available as a subscription provider. Connect your MiniMax account and Manifest automatically discovers your available models.

Routing scorer sees tools The scoring algorithm now knows when a request contains tools. Before, it only looked at the prompt text. A request with tools won’t be misclassified as “simple” anymore.

Bugfixes

• Skip prompt caching for Anthropic subscription auth (#1228) — Some models listed by OpenRouter don’t actually exist at the real providers. They were showing up in fallback chains and causing failures. Manifest now checks each model against the provider’s own API before including it. (#1224)
• Show correct costs when both API key and subscription are connected (#1230)
• Normalize Mistral tool-call IDs in OpenAI-compatible payloads (#1186)
• Prevent routing from falling back across auth types (#1185)
• Normalize DeepSeek max_tokens before forwarding (#1178)
• Switch Moonshot/Kimi API endpoint from .cn to .ai for international users (#1231)
• Fix page refresh returning 404 on deep routes in local mode (#1225)
• Fix /manifest command returning no visible reply in Telegram (#1077)
• Preserve Codex subscription instructions during routing (#1170)
• Trigger fallbacks on provider transport failures (#1181)
• Filter phantom OpenRouter models from fallback chains (#1224)

Performance

• Optimized API response times across 6+ endpoints to sub-500ms (#1197)
• Code splitting and lazy loading on the frontend (#1189)
• Skeleton loaders and spinner buttons for better perceived performance (#1189)

What’s coming next

• Expanding subscription model discovery to more providers

— The Manifest team