Manifest is an open‑source routing layer for large language model requests that scores each query locally in under 2 ms and forwards it to the most cost‑effective model in the user’s pool. It provides real‑time cost analytics, budget alerts, and OpenTelemetry‑compatible telemetry while keeping all data on the user’s infrastructure.
Funding
Funding not disclosed
Founders
Product
Problem
Organizations that rely on a single large language model for all requests incur unnecessary compute costs and expose all query data to a centralized service. The lack of granular model selection also prevents fine‑tuned cost control and makes budgeting unpredictable.
Solution
Manifest is an open‑source routing layer that intercepts each LLM request, evaluates its complexity locally in under 2 ms, and forwards it to the most cost‑effective model in the user’s pool. By performing query scoring on the client machine, the platform preserves data privacy and eliminates the need to send raw inputs to a third‑party proxy. Integrated cost analytics break down spend per message, allow users to set budget thresholds and receive alerts when limits are approached. The router emits telemetry in OpenTelemetry format, enabling seamless integration with existing observability stacks. A native OpenClaw plugin provides one‑command installation and automatic model discovery, so teams can adopt the router without code changes. Being fully open source, Manifest can be self‑hosted, inspected, or extended to support custom models and routing policies.
Target Audience
The primary users are AI engineers, DevOps teams, and product developers who run LLM workloads on OpenClaw and need granular cost control, privacy guarantees, and observability. It also serves enterprises and SaaS providers looking to optimize large‑scale language model usage across multiple applications.
Features
- Local query scoring engine (<2 ms latency) that classifies request complexity and selects the optimal model
- Multi‑model support with automatic fallback to smaller, cheaper models for simple tasks
- Real‑time cost breakdown per request, configurable budget limits, and alerting via webhook or dashboard
- OpenTelemetry‑compatible telemetry exporter for metrics, traces, and logs integration
- Native OpenClaw plugin offering single‑command deployment and zero‑code integration
- End‑to‑end data residency: all scoring and routing decisions occur on the user’s infrastructure
- Fully open‑source codebase with self‑hosting options and extensible routing policies