Skip to main content

What Are Sessions?

Sessions are where you interact with agents and watch them work in real-time. Each session is a complete conversation history between you and an agent, tracking every message, tool call, and decision. Think of sessions as chat conversations with full transparency. You see what the agent is thinking, what tools it’s using, and exactly how much each interaction costs.

Creating Sessions

Start a new session from the Sessions page by clicking “New Session”. You can:
  • Choose an agent - Select which agent to use, or let Celeria use your default agent
  • Name your session - Give it a descriptive name, or let the AI name it automatically after the first response
  • Start with a message - Jump right into conversation, or create an empty session for later
Sessions can also be created automatically when tasks execute. Each task run spawns a new session so you can observe what happened.

Session States

Sessions move through different states as they execute: ACTIVE - Ready to receive messages. The agent is waiting for your next input. STREAMING - Currently processing your message and responding. You’ll see text and tool calls appear in real-time as the agent works. PAUSED - You’ve paused execution mid-stream. The agent stops where it is, preserving its progress. Send another message to resume from that exact point. FAILED - Something went wrong during execution. ARCHIVED - Moved to history. Archived sessions don’t show in your active list but remain accessible for reference.

Real-Time Updates

Sessions update in real-time as agents work and as sessions are created, modified, or deleted across your workspace. When you send a message, Celeria streams the agent’s response as it happens:
  • Text content - The agent’s response appears word by word as it’s generated
  • Tool calls - When the agent uses integrations or core tools, you see the tool name, inputs, and results
  • Progress - Visual indicators show the agent is thinking, calling tools, or completing its response
Your session list stays current automatically. When new sessions are created (by you, teammates, or tasks), they appear immediately. Status changes, name updates, and cost calculations all sync in real-time across all your open tabs.

Pause and Resume

While an agent is streaming, you can pause execution at any time. Click the pause button and the agent stops immediately, preserving:
  • All messages generated so far
  • The current position in its reasoning
  • Token cache for seamless resumption
To resume, simply send another message. The agent picks up exactly where it left off, using cached tokens to avoid re-processing work it already completed. You can pause and resume as many times as needed. This is useful when you want to:
  • Stop an agent that’s going in the wrong direction
  • Review intermediate results before continuing
  • Adjust your next prompt based on what you’ve seen so far

Message History

Sessions store the complete conversation history, including:
  • User messages - What you asked or instructed
  • Assistant messages - The agent’s responses, including text and tool calls
  • Tool results - Outputs from integrations and core tools
  • Usage metadata - Token counts and costs for each turn
Messages are stored as structured data (JSONB) so they’re fast to query and filter. The full history is available whenever you open a session.

Cost Tracking

Every session tracks costs in real-time:
  • Session total - Running total displayed in the session header
  • Model-based pricing - Costs calculated using the selected model’s pricing (per million tokens)
Costs update as the session progresses, so you always know what you’re spending. This helps you:
  • Compare different models and agents
  • Track spending across tasks and workflows
  • Optimize prompts for cost efficiency

Smart Naming

Sessions need names so you can find them later. Celeria handles this intelligently: Auto-naming - After the first assistant response, Celeria uses AI to generate a 3-6 word title that summarizes the conversation. This happens automatically unless you’ve already named the session yourself. Manual naming - You can set the name when creating a session or rename it anytime from the session menu. Smart rename - Trigger AI-powered renaming manually through the context menu. Useful when a session’s focus has shifted. The naming system tracks whether a name is default, auto-generated, or user-modified, so it doesn’t overwrite your custom names.

Session Management

The Sessions page provides two views for managing your work: Grid View - Visual cards showing each session’s name, agent, status, cost, and last update. Quick access to common actions through the context menu. List View - Sortable table with columns for status, name, agent, cost, duration, and last update. Better for managing many sessions at once. Both views support:
  • Search - Filter sessions by name
  • Status filtering - Show only active, streaming, paused, or failed sessions
  • Agent filtering - View sessions for a specific agent
  • Sorting - Order by name, cost, duration, or update time
  • Context menus - Rename, duplicate, archive, delete, or smart rename
When you have a session open, the session list collapses to a sidebar. This lets you:
  • Switch between sessions quickly
  • See status indicators at a glance
  • Search and filter while working
  • Reorder sessions by dragging (your preferred order is preserved)
The sidebar shows a pulse animation on streaming sessions so you can monitor background work while focusing on another session.

Archiving and Restoring

Archive sessions when you’re done with them but want to keep the history. Archived sessions:
  • Move out of your active list
  • Remain accessible from the archive view
  • Preserve all messages and metadata
  • Can be restored to active status anytime
Restoring an archived session returns it to your active list with a new position at the end.

Changing Agents

You can change which agent a session uses through the session menu. This is useful when:
  • You want to hand off a conversation to a different agent
  • You need different integrations or tools for the next phase of work
  • You want to compare how different agents handle the same context
The session preserves its message history when you change agents. The new agent sees the full conversation and can pick up where the previous agent left off.

Duplicating Sessions

Duplicate a session to:
  • Try different approaches with the same starting context
  • Branch a conversation in a new direction while preserving the original
  • Reuse a conversation pattern as a template
The duplicate includes all messages up to the point of duplication but creates a new session with its own independent history going forward.

Session Context for Tasks

When a task executes, it creates a session automatically. This means:
  • Every task run is fully observable
  • You can see exactly what the agent did
  • Costs are tracked per execution
  • Failed task runs preserve error details
Task-created sessions appear in your session list with metadata linking them back to the task that spawned them.

Next Steps

  • Integrations - Give your sessions access to external services
  • Actions - Create reusable prompts for your agents to execute
  • Tasks - Set up autonomous agentic workflows