Hyperagents: Self-Referential Meta-Agents (2026)

Hyperagents is an open-source framework from Meta FAIR (facebookresearch/HyperAgents) that introduces self-referential agents capable of open-ended self-improvement. It unifies a task agent (solves the target task) and a meta agent (modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification that improves not only task-solving behavior but also the mechanism generating future improvements. Instantiating the Darwin Gödel Machine (DGM) as DGM-Hyperagents (DGM-H), it eliminates the need for domain-specific alignment between task performance and self-modification skill, supporting self-accelerating progress on any computable task. Validated on code generation, paper review, robotics, and Olympiad math, DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration. It also exhibits meta-level improvements such as persistent memory and performance tracking that transfer across domains and accumulate across runs. The project has garnered 2.1k likes on Hugging Face and is available under the MIT license.