Ax shows how to build RLM agents from DSPy signatures, without graphs or loops
dosco · x · 2026-07-25
The post argues that Ax agents can be built entirely from DSPy signatures.
From the image:
- the Ax RLM Agent is described as composed signatures
- public contracts map inputs → outputs
- pipelines can turn inputs and memories into a distilled context, then produce outputs
- runtime capabilities include querying an LLM, calling tools, discovering docs, recalling memories, and finalizing tasks
- optional context programs can checkpoint summaries, record tombstones, cache candidates, and manage operations
The main claim is that a relatively small signature-based setup can replace more complex graphs or loops, and still work well even with smaller models.
More from coding & agent
- Anaconda teams up with Kilo Code to bring coding agents into enterprise workflows — anacondainc · 2026-07-25
- OcularAudio MCP gives agents local YouTube transcripts and on-demand screenshots — Carbon-B · 2026-07-25
- A developer built a Colosseum arcade fighter with Claude Opus 5 — chrisfirst · 2026-07-25
- API Ninjas facts server adds a simple MCP tool for random facts — modelcontextprotocol · 2026-07-25
- Courier adds an MCP connector for notifications, templates, and integrations — modelcontextprotocol · 2026-07-25
- A step-by-step recipe from supervised learning to agentic world modeling — cwolferesearch · 2026-07-25