What's long-term memory
Long-term memory is the AI's ability to store and use information about you across all sessions. Open the app in a week — the agent remembers your previous conversation, your goals, your documents.
This is the opposite of «context window» — the default behavior of AI chats where the model only remembers the current dialogue. After closing the chat or starting a new one, AI forgets everything. You have to re-introduce yourself every time.
Why it matters
Real situations where long-term memory changes the experience:
- Nutrition — AI-nutritionist remembers your calorie target, allergies, last week's diary. Doesn't ask «what's your age» every time.
- Personal projects — a custom agent holds context of one project for weeks: stack, goals, deadlines, participants. Doesn't mix with other projects.
- Networking — AI-networker remembers every contact, when you met, what you discussed. Suggests follow-ups.
- Personal growth — AI-assistant sees your progress, typical struggles, what works and what doesn't. Advice becomes more precise.
Three memory layers
Memory in Synaps is hierarchical. Each layer is accessible to a specific slice of agents:
Profile
Visible to all agentsBasic info about you: name, allergies, weight-loss or language goals. What's shared across agents and doesn't change often. Set once.
Agent memory
Visible to one agent onlySpecialized context: food diary for nutritionist, legal cases for lawyer, contacts for networker. Isolated — other agents don't see this.
Conversation context
Current sessionShort-term buffer for current dialogue. After closing the chat, important facts migrate to agent memory or profile.
How it works technically
Under the hood, memory is a combination of two approaches:
- Structured records — key facts (name, goals, constraints) stored as a structured object in Postgres. Deterministic access — AI always knows your exact parameters.
- Embeddings + vector search — fragments of conversations, notes, uploaded documents are encoded as embeddings. On a new message, AI finds semantically close fragments and injects them as context.
Privacy and control
Your memory belongs to you:
- Stored on our EU/RU servers
- Not shared with third parties
- Not used for model training
- Editable and deletable via the app
- Full account deletion wipes everything
FAQ
How is Synaps memory different from ChatGPT Memory?+
ChatGPT Memory stores short «facts about the user» — fragments the model decides to keep. Synaps has structured multi-layer memory: goals, personal parameters, allergies, documents, history with a specific agent. Memory works across all devices (web, iOS, Android) without a subscription.
Do different agents share memory?+
No, each agent has its own isolated memory. Nutritionist doesn't know your legal cases, lawyer doesn't know your calories. This is an intentional architectural choice — each expert stays focused on its domain. Common profile data (name, basic goals) is shared.
Where is my memory physically stored?+
On our servers in the EU/RU regions. Only you have access — we don't share data with third parties, don't use it for training. Any dialogue or full profile can be deleted from the app at any moment.
Can I edit or wipe memory?+
Yes. The app has a Memory page — you can see all saved facts, edit or delete any. You can also fully reset memory of a specific agent, keeping only the profile.
Is it free?+
Yes. Long-term memory is included for all agents on all tiers. Free, with no caps on saved data volume.
Try AI with memory
30 seconds to sign up — AI remembers your context from the very first message. Free, no subscriptions.
Open Synaps AI →