Dagpipe

If you've ever watched an LLM pipeline crash three hours in and had to start over from scratch, this is the fix. DagPipe exposes operations for building multi-step AI workflows with automatic checkpoint recovery, where each node's output is validated and saved before moving forward. A failure at step 4 doesn't erase steps 1 through 3. Just rerun and it picks up where it stopped. The server integrates a model router that sends simple tasks to free-tier models and escalates complex ones, plus semantic assertions that catch structurally valid but logically wrong output. It's built for stdio transport and works with any MCP-compatible client. The checkpointing is filesystem-based by default, no database required, and the whole thing runs on free API tiers.