Explainer
Qu’est-ce que la mémoire des agents IA ? Court terme, long terme et contexte de projet
La mémoire d’un agent IA désigne les informations qu’il peut réutiliser dans les tâches suivantes au lieu de repartir de zéro. La mémoire à court terme sert au fil actuel, la mémoire à long terme conserve les faits durables et le contexte de projet relie les fichiers, décisions et livrables d’un même travail.
Les trois portées de mémoire
| Scope | Use |
|---|---|
| Short-term memory | Current thread, tool results, and resumable work |
| Long-term memory | Durable preferences, rules, and lessons across sessions |
| Project context | Sources, decisions, artifacts, and dependencies for one outcome |
Mémoire à court terme : le fil de travail
Short-term memory is the active state of a task. Keep recent decisions, unresolved questions, and tool results that affect the current deliverable. Summarize repetitive history and mark failed attempts so the agent does not repeat them blindly.
Mémoire à long terme : les faits durables
Long-term memory should hold only stable, useful, expected, and properly scoped facts. Give each memory an owner, timestamp, correction path, and retention rule. Do not turn a one-off request into a permanent preference or store secrets by default.
Contexte de projet : une mémoire délimitée
Project context keeps a brief, source files, approved facts, feedback, owners, and deadlines together. A project-specific claim should not silently become a global user preference. When a source changes, mark dependent work stale and show the provenance.
Récupération et gouvernance
Storing information is separate from retrieving it. Check scope, relevance, freshness, authority, and conflicts before adding a memory to the current context. Current instructions and approved project sources should normally outrank old inferences.
Échecs fréquents
- Remembering everything increases noise, cost, and privacy exposure.
- Wrong namespaces leak a project fact into unrelated work.
- Stale memories override current instructions without dates or precedence.
- Invisible memories cannot be corrected or deleted.
- Memory is evidence, not a guarantee of truth.
Le rôle d’Ottermind
Ottermind is an AI agent workspace that keeps goals, files, links, decisions, intermediate work, and follow-up tasks connected around their project. Use a bounded project, name the source of truth, review context before consequential output, and keep correction and forgetting available.
Questions fréquentes
Is chat history the same as agent memory?
No. Chat history is one source of short-term context. Agent memory also includes durable facts, summaries, checkpoints, and project records.
Does more memory make an agent smarter?
Not automatically. Better memory is scoped, fresh, traceable, editable, and relevant to the current task.
Conclusion
Short-term memory keeps a task moving, long-term memory carries carefully selected durable facts, and project context keeps one body of work coherent. Make the boundaries visible and design correction and forgetting from the start.
Research and disclosure: We reviewed the OpenAI Agents SDK memory guide, OpenAI Sessions guide, OpenAI context guide, and LangGraph memory documentation on August 27, 2026. These are implementation references, not an Ottermind benchmark.
