Sessions

Session Model

Sessions track conversation history and context for each agent. ThinkFleet uses a smart session system that maintains continuity across channels.

DM Sessions

By default, all your direct messages across all channels collapse into a single main session. This means your AI remembers context whether you message from WhatsApp, Slack, or Telegram.

You can change this behavior with session scope settings:

Scope	Behavior
Main (default)	All DMs share one session across channels
Per-peer	Separate session per contact
Per-channel-peer	Separate session per contact per channel

Group Sessions

Group conversations always use isolated sessions — one per group. This prevents group context from leaking into your private conversations.

Session format: agent:<agentId>:<channel>:group:<groupId>

Session Lifecycle

Daily reset — Sessions automatically reset at 4 AM (configurable) for freshness
Idle reset — Sessions can be configured to reset after a period of inactivity
Compaction — When sessions grow large (~100K tokens), they are automatically compacted to preserve the most relevant context
Memory flush — Before compaction, the agent is prompted to save important information to its long-term memory

Session Tools

Agents can interact with sessions programmatically:

List sessions — See all active sessions with filters
Inspect history — View transcript and token counts
Send to session — Cross-session communication
Spawn sub-tasks — Run isolated agent tasks from within a session