HyperAgent: Planning and Acting over Tool-Schema Hypergraphs for Tool-Use LLM Agents¶
Deterministic graph mechanism, not original external environments.
Paper information¶
| Field | Details |
|---|---|
| Paper | Paper or original source |
| Institution | University of New South Wales; first author Zian Zhai. |
| First public release | 2026-07-31 (arXiv v1) |
| Author code | No public author code found as of 2026-08-24. |
| Local adapter or method key | hyperagent |
| Local reproduction code | src/auto_research/agent_research/historical_b10_b11.py |
Paper method and evidence¶
Represent tools as input-schema→output-schema hyperedges. Construct a task DAG, then expand producer support for missing state.
flowchart LR
A[Represent tools as hyperedges] --> B[Build task DAG]
B[Build task DAG] --> C[Expand producer tools for missing state]Original paper figure¶
The original figure shows tool-schema hypergraph planning. From the original paper; copyright remains with the authors.
Core equation¶
\[
e_{tool}:\{schema_{in}\}\rightarrow\{schema_{out}\}
\]
Reported results¶
The paper reports higher AppWorld task completion with fewer redundant API calls, LLM interactions and tokens.
Local reproduction¶
PlanBench mini 120 episodes: HyperAgent 1.0000 joint success, mean cost 0.5500, versus long-context 1.0000 and 64.5000. Metrics, batch index. Audit traces when success saturates. No browser, AppWorld, SWE-bench, or external services. Raw runs stay in runs/agent-research/; only metrics committed.