Différences

Ci-dessous, les différences entre deux révisions de la page.

--- informatique:ai_lm:ai_agent [28/05/2026 13:14] – [OpenHands] cyrille
+++ informatique:ai_lm:ai_agent [22/06/2026 18:44] (Version actuelle) – [GBrain] cyrille
@@ Ligne 26: / Ligne 26: @@
 ===== À quoi ça sert ? =====
-J'ai mis du temps à me faire une idée sur l'utilité des agents ...
 ==== Agent de codage ====
@@ Ligne 111: / Ligne 110: @@
 https://nanobot.club/
+La home page parle de "Memory Management System, Hybrid Semantic Search, Algorithm: 70% Vector + 30% BM25" mais les ChatBots ne trouvent aucune trace de ces technologies dans le code source ... ¿¿ **//AIGen Marketing Bullshit//** ??
 ==== OpenHands ====
 OpenHands: AI-Driven Development
+OpenDevin = OpenHands
   * https://www.openhands.dev/
@@ Ligne 140: / Ligne 142: @@
   * [[https://marketplace.visualstudio.com/items?itemName=Poppywu124.hermes-chat|Hermes Agent Chat]] (2026-04-27: 1399 installs)
+===== Memory systems =====
+==== Basic Memory ====
+  * https://basicmemory.com/
+  * https://github.com/basicmachines-co/basic-memory
+Plain Markdown files, semantic connections, and a knowledge graph that grows with every conversation.
+  * Local-first. Plain text on your disk. Forever.
+  * Two-way. AI and humans write to the same text (markdown) files; sync keeps them in step.
+    * indexing with SQLite/PostgreSQL
+  * A real knowledge graph. Observations and wikilinks compound into context.
+  * Semantic search. Find notes by meaning, not just keywords.
+    * [[https://github.com/qdrant/fastembed|Fastembed]]
+  * MCP-native. Works with every major AI client and IDE.
+  * Progressive tool discovery. Every tool is tagged with behavior hints (read-only, destructive, idempotent) so agents pick the right tool on demand — no wasted context trying things to see what they do.
+  * Cloud, optional. Sync across devices when you want — never required.
+==== GBrain ====
+https://github.com/garrytan/gbrain
+Semble le plus aboutit, mais c'est une sacrée usine !
+relations automatique
+  * GBrain extrait automatiquement les relations typées avec des RegEx : works_at, founded, attended, invested_in, advises
+  * On peut [[https://github.com/garrytan/gbrain/blob/master/docs/schema-author-tutorial.md|définir/étendre le schema]] mais ce n'est pas trivial. 😥
+  * Avec un fallback LLM en option
+Pour configurer un model embenddings custom il faut utiliser la clé "provider_base_urls.ollama" :
+<code>
+$ cat ~/.gbrain/config.json
+{
+  "embedding_model": "ollama:tei",
+  "embedding_dimensions": 768,
+  "provider_base_urls": {
+    "ollama": "http://127.0.0.1:9000/v1"
+  },
+  ...
+</code>
+Puis ''gbrain init --pglite --embedding-model ollama:tei --embedding-dimensions 768''
+==== Mem0 ====
+https://mem0.ai/
+AI memory that persists across sessions and agents.
+Trop lourd pour un usage perso ...
+==== Stash Memory ====
+https://alash3al.github.io/stash/
 ===== Frameworks =====
@@ Ligne 193: / Ligne 251: @@
   * Choose DSPy for: minimal boilerplate, readable single-file flows, contract-driven development (with optional optimizers).
   * could show dramatically different results when using its signature optimizers (like MIPROv2) and Chain-of-Thought prompting, which can significantly improve answer quality.
+===== No-code frameworks =====
+==== Flowise ====
+  * https://flowiseai.com/
+  * https://github.com/FlowiseAI/Flowise