Resume a Session

Continue a previous conversation with full model/agent/history restored

Resume a previous conversation exactly where you left off — same model, same agent, same chat history.

Quick Start

1. Find your session ID

   praisonai session list

   Copy the session ID from the output table.

2. Or skip the id — pick from a list

   On an interactive terminal, omit the id to open a picker of recent sessions:

   praisonai session resume

   Select a session to resume
   #   Title           Updated               Messages   Tokens    Cost
   1   fix-auth-bug    2026-09-22T14:31:07   12         12,440    $0.0710
   2   draft-release   2026-09-22T11:02:44   5          3,120     $0.0180
   
   Session number (0 to cancel):

   Type a number to resume that session, or 0 to cancel. See Interactive Picker.

3. Resume the session

   praisonai session resume <session_id>

   The CLI restores your prior model, agent name, full chat history, and cumulative usage totals, then shows:

   ╭─ Session Resumed ──────────────────────────────────╮
   │ Session: my-assistant                               │
   │ Model: anthropic/claude-3-5-sonnet-latest           │
   │ Messages restored: 12                               │
   │ Usage:  12,440 tokens · $0.0710 (3 requests)        │
   ╰─────────────────────────────────────────────────────╯
   
   --- Restored Conversation ---
   [user] What is the capital of France?
   [assistant] Paris.

   The Usage line appears only when token/cost data has been accumulated for the session. Cumulative totals keep accumulating from this point — they do not reset.

4. Continue with a new prompt

   praisonai session resume <session_id> "What about Germany?"

   Passes the resumed state directly into the same run path as praisonai run --session.

5. Give the session a friendly title (optional)

   Rename the session so it’s easy to spot in session list — the id you resume with never changes:

   praisonai session rename <session_id> "fix-auth-bug"

   The title then surfaces in the Name column of session list. See Session Rename for details.

6. View transcript only (read-only)

   praisonai session resume <session_id> --transcript

   Prints recent events without restoring state — useful for reviewing a session without continuing it.

Resume into the interactive TUI

The same resume primitive is available on the bare praisonai interactive launch — no need to type /continue inside a cold TUI. As of PR #4943, the interactive TUI auto-persists every conversation, so this primitive now has data to resume against on a fresh install — you no longer need to have previously passed --auto-save:

# Most recent session
praisonai -c

# Specific session id (equivalent to praisonai session resume <id> without a follow-up prompt)
praisonai --session <id>

# Fork non-destructively
praisonai --fork --session <id>

The recorded session model is restored, the prior transcript is replayed into scrollback, and the primary agent is wired to the resumed session’s store so your next prompt continues the conversation with full memory of prior turns.

Added in PR #4912 as parity with run --continue / chat --continue. See Interactive TUI → Launching with a resumed session.

Interactive Picker

Omit the session id on an interactive terminal and the CLI opens a picker of recent sessions — no more copy-pasting an opaque id.

The picker fires only when all of these hold: a real TTY, no --json, and no id passed. It lists the same project-scoped sessions as session list (most-recently-updated first, up to 20).

Select a session to resume
#   Title           Updated               Messages   Tokens    Cost
1   fix-auth-bug    2026-09-22T14:31:07   12         12,440    $0.0710
2   draft-release   2026-09-22T11:02:44   5          3,120     $0.0180

Session number (0 to cancel):

Enter a number 1..N to resume that session, or 0 to cancel. Ctrl-C / EOF also cancels.

When it exits without resuming

The picker never silently starts a fresh run. Three cases exit without resuming:

Non-TTY / --json

Scripts and CI are unaffected — the id stays required. With no id under --json or a non-TTY, the command errors instead of prompting:

Error: Missing session id
  → Pass a session id, or run interactively to pick one

How It Works

The rehydrate_session() helper searches stores in order and returns a RehydratedSession dataclass:

from dataclasses import dataclass, field
from typing import Optional

@dataclass
class RehydratedSession:
    session_id: str
    chat_history: list[dict[str, str]] = field(default_factory=list)
    model: Optional[str] = None
    agent_name: Optional[str] = None
    metadata: dict = field(default_factory=dict)
    usage: dict = field(default_factory=dict)  # cumulative token/cost totals
    found: bool = False

When found is False, the CLI prints:

Error: Session not found: <session_id>
  → Use 'praisonai session list' to see available sessions

Model Restoration

Resuming restores the model the session was running with — not the current default. Change your default model between runs and a resumed session still runs on the original model.

When the recorded model is used, the CLI prints:

Restored session model: anthropic/claude-3-5-sonnet-latest

Precedence (first match wins): explicit --model → config agent.model → recorded session model → keyless local-first fallback → default resolution.

Override the recorded model

Pass --model on resume — it wins and becomes the new recorded model for subsequent turns.

praisonai session resume abc123 --model gpt-4o "continue with a different model"

Where the model comes from

The store resolves the recorded model in this order: the session-level metadata["model"] (written by the session-continuity path), then the model on the most recent turn that recorded one. When neither exists, restoration is skipped and normal default resolution applies.

Legacy sessions

Sessions written by older PraisonAI versions stored model / llm as top-level fields rather than under metadata. The store folds those legacy fields back into metadata on load, so historical sessions recover their recorded model on resume without any migration step. Existing metadata values always win over the mirrored top-level value.

Cumulative Usage Across Resumes

Token and cost totals accumulate across all runs for a session. When you resume and run more prompts, the totals grow — they are never reset.

The store that already holds usage metadata takes priority on resume — a globally-stored session resumed via the project store keeps its original totals intact.

Project vs Global Store

The resume helper always searches the project-scoped store first, then falls back to the global default store. This means sessions created inside a project are resolved with higher priority, and you can still resume cross-project sessions by ID.

The project_path parameter (optional, defaults to cwd) controls which project store is searched first. Use it when running resume from a different working directory:

praisonai session resume <id>

Flags

Relation to run --continue / run --session

praisonai session resume and praisonai run --session <id> call the same rehydrate_session() helper under the hood. The outcomes are identical: same chat history, same model, same agent. Use whichever form fits your workflow.

# These are equivalent:
praisonai session resume abc123 "next question"
praisonai run --session abc123 "next question"

As of the fix for PraisonAI #2655, run --continue also searches both stores when finding the last session — before, only session resume did.

As of praisonai/PraisonAI#3203, both the resume path and the show/delete/export path enumerate the same store instances (same classes, same order) by delegating to a single canonical_cli_stores() helper — so an id you can resume is guaranteed to also be show-able, delete-able, and export-able.

Best Practices

SESSION_ID="abc-123-def"
praisonai session resume "$SESSION_ID" "run daily report"

Related