# Continuum

A Python framework for building and running AI agents at scale, combining a typed agent core with cost-aware model routing, persistent memory, MCP-native tool calling and durable, resumable workflows.

Continuum is a Python framework, from ShyftLabs, for building, orchestrating and deploying AI agents. It brings together a strongly typed agent core with full lifecycle hooks and schema-validated structured output, nine composable multi-agent patterns such as sequential, parallel, loop, routing, planning, reflection, debate and supervised execution, and a small, type-safe API over all of it.

A cost-aware inference router classifies each request by complexity and sends it to the cheapest model that can handle it, with automatic failover across providers and no lock-in to a single one. Stateful memory combines persistent long-term recall with low-latency working memory, with multi-tenant isolation and built-in personal-data redaction, and tools are called through the Model Context Protocol across multiple transports with fine-grained scoping. Durable execution keeps long-running workflows crash- and restart-safe, with human approval gates, and distributed tracing and token, latency and error telemetry give full visibility into a run.

It targets both frontier and open-weight models through a single model identifier string, so switching providers does not require touching agent code.

Website: https://docs.continuum.shyftlabs.io/