Managed CLI - PraisonAI

The praisonai managed command group lists and reclaims managed sandboxes across every provider (docker, e2b, modal, daytona, flyio, tenki) and manages Anthropic-hosted sessions, agents, and environments from the terminal.

from praisonaiagents import Agent

agent = Agent(name="managed-agent", instructions="Use managed CLI commands for automation.")
agent.start("Run the managed deploy command for production.")

The user runs praisonai managed commands to create, inspect, and tear down hosted agents and sessions from the terminal.

Quick Start

Set API Key

Export your Anthropic API key:

export ANTHROPIC_API_KEY=sk-ant-...

List Sessions

View sessions for an agent:

praisonai managed sessions list agent_01AbCdEf

Delete with Confirmation

Clean up resources safely:

praisonai managed sessions delete sesn_01AbCdEf --yes

How It Works

Commands

Sessions

Manage Anthropic-hosted agent sessions:

Command	Purpose	Example
`sessions list <agent_id>`	List sessions for an agent	`praisonai managed sessions list agent_01AbCdEf`
`sessions get <session_id>`	Get session details and usage	`praisonai managed sessions get sesn_01AbCdEf`
`sessions resume <session_id> "<prompt>"`	Resume session with new prompt	`praisonai managed sessions resume sesn_01AbCdEf "Continue the task"`
`sessions delete <session_id>`	Delete a session (with confirmation)	`praisonai managed sessions delete sesn_01AbCdEf --yes`

# List recent sessions
praisonai managed sessions list agent_01AbCdEf --limit 5

# Get detailed session info
praisonai managed sessions get sesn_01AbCdEf

# Resume previous conversation
praisonai managed sessions resume sesn_01AbCdEf "What did we discuss?"

# Clean up old sessions
praisonai managed sessions delete sesn_01AbCdEf --yes

Agents

Manage your Anthropic-hosted agents:

Command	Purpose	Example
`agents list`	List all your agents	`praisonai managed agents list`
`agents get <agent_id>`	Get agent configuration	`praisonai managed agents get agent_01AbCdEf`
`agents create`	Create new agent (via Python SDK)	See HostedAgent docs
`agents update <agent_id>`	Update agent name, system, or model	`praisonai managed agents update agent_01AbCdEf --name "New Name" --version 1`
`agents delete <agent_id>`	Delete an agent (with confirmation)	`praisonai managed agents delete agent_01AbCdEf --yes`

# List your agents
praisonai managed agents list

# View agent details
praisonai managed agents get agent_01AbCdEf

# Update agent configuration  
praisonai managed agents update agent_01AbCdEf --system "You are a data analyst." --version 2

# Remove unused agent
praisonai managed agents delete agent_01AbCdEf --yes

Environments

Manage sandbox environments for code execution:

Command	Purpose	Example
`envs list`	List your environments	`praisonai managed envs list`
`envs get <env_id>`	Get environment details	`praisonai managed envs get env_01AbCdEf`
`envs update <env_id>`	Update packages or networking	`praisonai managed envs update env_01AbCdEf --packages "pandas,numpy" --networking full`
`envs delete <env_id>`	Delete an environment (with confirmation)	`praisonai managed envs delete env_01AbCdEf --yes`

# List environments
praisonai managed envs list

# Install packages
praisonai managed envs update env_01AbCdEf --packages "requests,beautifulsoup4"

# Change networking mode
praisonai managed envs update env_01AbCdEf --networking limited

# Clean up environment
praisonai managed envs delete env_01AbCdEf --yes

Environment Update Options:

--packages: Comma-separated pip packages (whitespace is trimmed, empty entries filtered)
--networking: Either full or limited (case-sensitive)

Sandbox Lifecycle: ps / stop

List and reclaim running agent sandboxes across every provider.

Command	Purpose	Example
`managed ps`	List every running agent sandbox across all providers	`praisonai managed ps`
`managed ps --json`	Machine-readable output (`{"sandboxes": [...], "errors": [...]}`)	`praisonai managed ps --json`
`managed ps --provider docker` (or `-p`)	Filter to one provider	`praisonai managed ps -p docker`
`managed stop <instance-id>`	Reclaim one sandbox by ID	`praisonai managed stop docker_a1b2c3d4`
`managed stop --all`	Reclaim every running sandbox (across providers)	`praisonai managed stop --all`
`managed stop --all --provider docker`	Reclaim every sandbox from one provider	`praisonai managed stop --all -p docker`

Sandboxes normally self-destruct when a run ends. A crashed or killed run leaves one behind, and ps / stop are the way to see and reclaim it.

INSTANCE ID                PROVIDER   STATUS     UPTIME   IMAGE
------------------------------------------------------------------------------
docker_5afaa3bfa704        docker     running    42s      python:3.12-slim

1 running. Stop with: praisonai managed stop <instance-id>

Behaviour:

Providers that are not installed or not configured (no E2B key, Docker daemon down, etc.) are skipped silently — absence of credentials is a normal state, not an error.
An explicit --provider foo for an uninstalled or unknown provider is surfaced as an error (exit code 1) — you asked for it, so silence would mislead.
Available providers whose lookup fails are surfaced under errors: in the JSON output and printed at the bottom of the text output; the exit code is 1 in that case, so ps / stop never report a clean sweep when some providers could not be queried.
Docker containers are discovered cross-process via the praisonai=managed label and praisonai_<id> naming convention, so containers started by an earlier CLI or a crashed script are still visible from a fresh CLI invocation.
ps and stop scan docker, e2b, modal, daytona, flyio, and tenki. local is excluded — it runs on this machine, so there is nothing to reclaim.
Uptime is formatted as Ns under a minute, Nm under an hour, NhMMm under a day, and NdHHh beyond that.

IDs Helper

Save and restore Anthropic-assigned resource IDs locally:

Command	Purpose	Example
`ids save <agent_id> <env_id> <session_id>`	Save IDs to JSON file	`praisonai managed ids save agent_01... env_01... sesn_01... --version 1`
`ids restore "<prompt>"`	Restore IDs and run prompt	`praisonai managed ids restore "Continue previous task"`
`ids show`	Display saved IDs	`praisonai managed ids show`

# Save current resource IDs
praisonai managed ids save agent_01AbCdEf env_01AbCdEf sesn_01AbCdEf --version 1

# View saved IDs
praisonai managed ids show

# Resume with saved state
praisonai managed ids restore "What was I working on?"

Choosing a Subcommand

Safety: Confirmation Prompts

All delete commands prompt for confirmation unless --yes is provided:

# Shows prompt: "Are you sure you want to delete sesn_01AbCdEf?"
praisonai managed sessions delete sesn_01AbCdEf

# Skips prompt
praisonai managed sessions delete sesn_01AbCdEf --yes

Typing anything non-affirmative prints Deletion cancelled. and exits with code 0.

Authentication

The CLI requires the anthropic Python package and one of these environment variables:

# Primary (checked first)
export ANTHROPIC_API_KEY=sk-ant-api-...

# Fallback
export CLAUDE_API_KEY=sk-ant-api-...

Installation:

pip install 'anthropic>=0.94.0'

Without the package or API key, commands exit with code 1 and an error message.

Common Patterns

Cleanup Script

Remove old sessions and unused resources:

#!/bin/bash
# cleanup-resources.sh

AGENT_ID="agent_01AbCdEf"

# List and delete old sessions
echo "Cleaning sessions for $AGENT_ID..."
praisonai managed sessions list $AGENT_ID --limit 20 | tail -n +3 | while read id status title; do
  if [[ $status == "completed" ]]; then
    praisonai managed sessions delete $id --yes
    echo "Deleted session: $id"
  fi
done

# List and delete unused environments  
echo "Cleaning environments..."
praisonai managed envs list | tail -n +3 | while read id status name; do
  if [[ $status == "inactive" ]]; then
    praisonai managed envs delete $id --yes
    echo "Deleted environment: $id"
  fi
done

Environment Setup

Install packages and configure networking:

# Update development environment
ENV_ID="env_01AbCdEf"
praisonai managed envs update $ENV_ID \
  --packages "pandas,numpy,matplotlib,jupyter" \
  --networking full

# Verify installation
praisonai managed envs get $ENV_ID

Session Backup

Save session state before major changes:

# Backup current session state
praisonai managed ids save agent_01AbCdEf env_01AbCdEf sesn_01AbCdEf \
  --version 1 --file backup-$(date +%Y%m%d).json

# Show what was backed up
praisonai managed ids show --file backup-$(date +%Y%m%d).json

Recover Stray Sandboxes

Find sandboxes left behind by a crashed run and reclaim them:

# See what is still running
praisonai managed ps

# Reclaim one by ID
praisonai managed stop docker_5afaa3bfa704

# Or reclaim everything at once
praisonai managed stop --all

Best Practices

Resource Management

Use --yes in scripts to avoid interactive prompts
Regularly clean up completed sessions to manage costs
Save important session IDs before deleting resources
Monitor environment status before making updates

Error Handling

Check exit codes in scripts (0 = success, 1 = error)
Handle API timeouts with retry logic
Verify API key is set before running commands
Use --limit to prevent large resource listings

Environment Updates

Test package installations in development environments first
Use --networking limited for security-sensitive workloads
Keep package lists minimal to reduce startup time
Update packages gradually to identify compatibility issues

Session Continuity

Use sessions resume instead of creating new sessions
Save session IDs for important conversations
Monitor session usage with sessions get
Implement session rotation for long-running applications

Hosted Agent

Python SDK for Anthropic-hosted agents

HostedAgent + DB persistence

Library-level persistence with database backends

Sessions

Advanced session management concepts

Shared Sandbox

Share one sandbox across a team or workflow with run_on=

​Quick Start

​How It Works

​Commands

​Sessions

​Agents

​Environments

​Sandbox Lifecycle: ps / stop

​IDs Helper

​Choosing a Subcommand

​Safety: Confirmation Prompts

​Authentication

​Common Patterns

​Cleanup Script

​Environment Setup

​Session Backup

​Recover Stray Sandboxes

​Best Practices

​Related

Hosted Agent

HostedAgent + DB persistence

Sessions

Shared Sandbox

Quick Start

How It Works

Commands

Sessions

Agents

Environments

Sandbox Lifecycle: ps / stop

IDs Helper

Choosing a Subcommand

Safety: Confirmation Prompts

Authentication

Common Patterns

Cleanup Script

Environment Setup

Session Backup

Recover Stray Sandboxes

Best Practices

Related