https://grainworks.tech/tags/agents/Recent content in Agents on grainworksHugoen-USWed, 10 Jun 2026 16:30:00 -0400https://grainworks.tech/projects/memory-research/Wed, 10 Jun 2026 16:30:00 -0400https://grainworks.tech/projects/memory-research/<p>A deep-dive comparison of two SQLite-backed agentic memory systems for Hermes Agent: <strong>Sibyl Memory</strong> (the #2-ranked system on LongMemEval) and <strong>Holographic</strong> (our local-first baseline).</p> <h2 id="the-question">The Question</h2> <p>Agents need memory to work across sessions. But every memory system makes a tradeoff: cloud vs local, free vs paid, simple vs feature-rich. Which one is right for an offline-first, zero-tracking research setup?</p> <h2 id="key-findings">Key Findings</h2> <table> <thead> <tr> <th>Dimension</th> <th>Holographic</th> <th>Sibyl Memory</th> </tr> </thead> <tbody> <tr> <td>Storage</td> <td>SQLite + FTS5 (3 tables)</td> <td>SQLite + FTS5 (5 tiers)</td> </tr> <tr> <td>Locality</td> <td>Fully offline</td> <td>Requires cloud activation</td> </tr> <tr> <td>Cost</td> <td>Free, unlimited</td> <td>Free cap: 2 MB</td> </tr> <tr> <td>Feature</td> <td>Trust scoring, HRR vectors</td> <td>Self-learning, 5-tier schema</td> </tr> <tr> <td>Search</td> <td>FTS5 + trust-weighted</td> <td>FTS5 + multi-tier + LLM re-ranking</td> </tr> <tr> <td>Dependencies</td> <td>numpy (optional)</td> <td>pip + cloud account</td> </tr> </tbody> </table> <h2 id="the-verdict">The Verdict</h2> <ul> <li><strong>Sibyl wins</strong> on features — self-learning, multi-tier schema, validated retrieval. But the 2 MB free cap and cloud activation requirement are dealbreakers for an offline-first system.</li> <li><strong>Holographic wins</strong> on operational simplicity — zero dependencies, unlimited storage, no accounts. Less flashy but operationally bulletproof.</li> </ul> <h2 id="where-we-landed">Where We Landed</h2> <p>We&rsquo;re staying on Holographic for the orchestrator, research pipelines, and autonomous morning briefing. The self-learning gap is real, but not worth trading local-first operation.</p>https://grainworks.tech/projects/sibyl-memory-evaluation/Mon, 08 Jun 2026 16:00:00 -0400https://grainworks.tech/projects/sibyl-memory-evaluation/<p><strong>Status: Published. Decision: staying on Holographic for now.</strong></p> <p>A head-to-head comparison of two SQLite-backed agentic memory systems for Hermes Agent:</p> <ul> <li><strong>Holographic:</strong> Local-first, zero-dependency memory provider with Phase Holographic Reduced Representations (HRR) for compositional retrieval. Unlimited database size, no accounts, no cloud roundtrips.</li> <li><strong>Sibyl Memory:</strong> Five-tier hierarchical schema (HOT/WARM/COLD/REFERENCE/ARCHIVE) with self-learning, LongMemEval-validated (95.6%), but requires cloud activation and has a 2 MB free cap.</li> </ul> <h2 id="the-verdict">The Verdict</h2> <p>Sibyl wins on features and benchmarks. Holographic wins on operational simplicity — zero external dependencies, unlimited storage, trust-based fact management.</p>