[Submitted on 18 Sep 2026 (v1), last revised 18 Sep 2026 (this version, v4)]
Title:ACE: JSON Tuple Spaces for Agent Coordination
Abstract: David Gelernter’s tuple-space model for Linda lets processes communicate through tuples that exist independently of their producers and are retrieved by content. The Agent Coordination Engine (ACE) extends this model with JSON records, access lists, expiration, bounded waiting, and optional model-based matching. These extensions support software agents that assign work, share results, and retry tasks as participants change. This report relates ACE’s operations to Linda’s, develops motivating scenarios, and specifies exact matching over JSON paths, embeddings and language-model predicates, ordered retrieval, and deletion confirmation within a visibility interval. Source inspection and the existing test suite support the account of the Go and SQLite implementation. The analysis identifies the conditions under which task claims exclude competing consumers, the application work required to handle retries, and the limits of cursor iteration, model-based reads, and identity enforcement across interfaces.
| Comments: | 9 pages, 2 tables |
| Subjects: | Distributed, Parallel, and Cluster Computing (cs.DC) |
| Cite as: | marXiv:2609.00008 [cs.DC] (or marXiv:2609.00008v4 [cs.DC] for this version) |
Submission history
[v1] Fri, 18 Sep 2026 15:15:42 UTC (268 KB)
[v2] Fri, 18 Sep 2026 15:19:14 UTC (268 KB)
[v3] Fri, 18 Sep 2026 15:46:14 UTC (268 KB)
[v4] Fri, 18 Sep 2026 15:50:47 UTC (267 KB)