Engraphis.com

08/05/2026

Give your agent more room to think. https://engraphis.com/

We measured how much Engraphis saves across long histories, retrieval, responses, and memory cleanup:

• Long project history sent to the model: 98% fewer tokens (49.9M → 891K on a 10-conversation, 1,986-question diagnostic)
• Retrieved memory content per question: 73.0% less (808.8 → 218.4 tokens), same Recall@5 of 1.000
• Smallest useful memory returned: 73% less (162.2 → 42.4 tokens)
• Complete memory-tool response: 55% less (17,172 → 7,663 tokens)
• Repeated memories after consolidation: 47% less (230 → 120 tokens), originals stay auditable

Same 512-token budget = 53x more evidence than recency-only retrieval. Including indexing, that's 97% less total token usage.

This is what focused, inspectable memory looks like for coding agents.

07/31/2026

Big update for Engraphis: agent memory recall is now much more efficient, using far less context and fewer tokens per query while staying grounded in the same facts.

Local-first memory engine for AI coding agents, stored right on your machine with a full dashboard to see how it all connects. Free and open source.

https://engraphis.com/

07/28/2026

AI systems lose context when memory stays scattered. Engraphis turns long-term AI memory into a living, visual knowledge graph, so you can see how ideas, entities, and causes connect.

In this short demo, watch an AI memory graph come alive and explore a clearer way to organize context for agents, apps, and teams.

Follow along here or on https://engraphis.com/ for practical ideas on AI memory, agent infrastructure, and connected knowledge.

07/27/2026

Context engineering is memory policy in practice. An agent needs scoped facts, expiry, provenance, and a correction path so the context it carries into the next run stays useful. That is the difference between durable memory and a growing pile of stale notes.

07/25/2026

Most AI coding agents are amnesiacs. Every new session, they re-read your whole repo from scratch and re-learn the same conventions, gotchas, and decisions you already explained last week. That's not intelligence, that's expensive repetition. Memory is the missing layer.

07/24/2026

AI agent memory is having a moment. Layered memory engines, temporal knowledge graphs, and persistent memory surveys are showing up everywhere this week — and for good reason. Agents that forget everything between sessions (or worse, trust every memory entry blindly) don't scale and aren't secure.

Engraphis takes a local-first, inspectable approach built specifically for coding agents: bi-temporal history so you can see what changed and when, code-aware recall, and MCP support out of the box. Check it out at engraphis.com.

07/24/2026

Engraphis 1.0.0 is out. Local-first memory for coding agents, now hardened for production.

What shipped: repo-wide CodeQL gate, zero security findings required to merge, SQLite storage, MCP native with 29 tools.

Apache 2.0.

Local-first, inspectable memory for coding agents: durable context across sessions and repositories, code-aware recall, bi-temporal history, MCP, and a self-hosted WebUI. - Coding-Dev-Tools/engraphis