Logo SVG copied to clipboard
Early Access Build agents that never fail Learn more

Manifest Weekly #08

Manifest Weekly #08 cover

The dashboard now shows how much money routing saves you.


What we shipped this week

See how much routing saves you: The Overview dashboard now shows a “Saved cost” metric. It compares what you actually spent against what you would have spent using a single model for everything.

Easier Hermes onboarding: Hermes users can now set up Manifest directly from hermes model with step-by-step instructions in the dashboard.


Fixes

  • Stalled streaming responses now trigger the fallback chain instead of hanging until timeout (#1657)
  • Streaming requests to some providers were showing 0 tokens and no cost (#1758)
  • Agents wrapping the user prompt in metadata were throwing off the complexity scorer, routing too many requests to expensive models (#1770)
  • OpenClaw setup snippet was pointing at the wrong API, causing empty chat bubbles (#1751)
  • Routing from a reasoning model to Mistral failed because reasoning_details was being forwarded (#1761)
  • Deleting an agent and recreating it with the same name now starts fresh instead of conflicting (#1776)
  • Manifest now detects Podman and Kubernetes as self-hosted environments (#1785)

Community contributions

Thanks to @el-analista for streaming fallback on stalled providers, @brian-c11 for making provider timeout configurable via PROVIDER_TIMEOUT_MS, and @danielbnelson for fixing blank pages in Safari on local installs.


What’s coming next

  • Add Groq provider
  • Allow users to benchmark models
  • Expose an Anthropic-compatible endpoint to support Anthropic-SDK agents.

— The Manifest team

Don't let your requests fail again and again

Manifest fixes your bad LLM requests on the fly so they return successful responses before it reaches your agent. No downtime.

  • Deprecated / Not-found models
  • Wrong parameters
  • Malformed requests
  • Exceeded context windows
Book a demo or