Manifest Weekly #01

Last week: subscriptions, fallback, and a bug fixes

Use your Claude Max/Pro subscription with Manifes

We shipped Anthropic OAuth support. If you have a Claude Max or Pro subscription, you can now connect it directly to Manifest and route all your AI traffic through your existing plan at zero additional cost. No API key needed. Just log in with your Anthropic account, and Manifest automatically discovers which models your subscription gives you access to. Teams already paying for Claude can plug Manifest in as their AI gateway without spending an extra dollar on API credits.

Automatic model fallback

If a model hits its rate limit or goes down, Manifest now automatically retries on the next model in your fallback chain. You configure up to 5 fallback models per routing tier from the dashboard, drag and drop to set the order. For example, if Claude Sonnet 4 is rate-limited, it can fall back to Claude Haiku, or to GPT-4.1, or to Gemini. Any model, any provider. Your users never see a failure. We also redesigned the routing page to make configuring all of this more intuitive.

Dynamic model discovery

We replaced the old model list with live discovery from each provider’s native API. When you connect a provider, Manifest fetches the actual models available to your account. No more outdated hardcoded lists. Pricing is synced from OpenRouter’s public API daily. Subscription models get separate discovery so you always see the right models for your plan.

Major bugfixes

• The embedded server never started in local mode due to a lifecycle bug in the plugin system.
• Pages returned 404 on reload in local mode.
• Anthropic and Google adapters dropped tool calls in streamed responses.
• The proxy crashed when all fallbacks were exhausted, causing infinite client retries.
• The agent list didn’t refresh after deleting an agent.
• DeepSeek reasoning content was rejected by other providers during cross-provider routing.

Performance

• Optimized 6+ slow API endpoints to sub-500ms response times
• Code splitting and lazy loading on the frontend
• Skeleton loaders and spinner buttons for better perceived performance
• Dropped redundant database indexes, reduced connection pool size

This week

• Expand subscription model discovery to more providers