🐾

Welcome to PocketPaw

Enter your access token to continue.

Run docker exec pocketpaw cat /home/pocketpaw/.pocketpaw/access_token to get your token.

Thinking... Preparing response... ▌
PocketPaw

No activity yet

Tool calls and agent events will appear here

No logs yet

System and agent events will appear here

Loading Crew...
Live

No tasks yet

Create a task to get your agents working

Task
Time
Actions

Agents

No agents yet

Activity

No activity yet

Select a project to view details

Task Details

Working
Live Output
Started · chunks
Elapsed:

Project Details

Add Task to Project

New Task

New Agent

Start a Project

Settings

Validation warnings (settings were saved)

General

is not installed.

This backend requires an external tool:

Documentation

Also requires an external CLI:

API Key Required

Uses Claude Code CLI with Anthropic API. No API key needed if logged into Claude Code. Runs via Ollama's Anthropic Messages API compatibility (Claude Code v0.14.0+). Use any model on OpenRouter with full CLI tool access via their Anthropic-compatible API. Connect to any Anthropic-compatible endpoint (LiteLLM, vLLM, etc.). Route through LiteLLM proxy for HuggingFace, vLLM, Together AI, Groq, and 100+ providers.
Run ollama list to see available models.
Get your key at openrouter.ai/keys
Browse models at openrouter.ai/models
Use provider/model format (e.g. huggingface/meta-llama/Llama-3-70b)
Set to match your model's limit (e.g. 8192 for DeepSeek). 0 = provider default.
Leave empty to let Claude Code auto-select the best model
Max agent tool-use loops per query (default: 25)
Uses Ollama's OpenAI-compatible API. No API key needed. Use any model on OpenRouter. Free and paid models available. Connect to any OpenAI-compatible endpoint (NVIDIA NIM, vLLM, etc.). Route through LiteLLM proxy for HuggingFace, vLLM, Together AI, Groq, and 100+ providers.
Get your key at openrouter.ai/keys
Browse models at openrouter.ai/models
Use provider/model format (e.g. huggingface/meta-llama/Llama-3-70b)
Set to match your model's limit (e.g. 8192 for DeepSeek). 0 = provider default.

Requires Google ADK. Install with: pip install 'pocketpaw[google-adk]'

Route through LiteLLM proxy. Use provider/model format (e.g. openai/gpt-4o).
Use provider/model format (e.g. huggingface/meta-llama/Llama-3-70b)
Set to match your model's limit (e.g. 8192 for DeepSeek). 0 = provider default.

Requires Codex CLI installed. Install with: npm install -g @openai/codex

Requires copilot CLI + pip install github-copilot-sdk

Requires pip install deepagents. Uses LangChain's init_chat_model for multi-provider support.

Format: provider:model. Examples: anthropic:claude-sonnet-4-6, openai:gpt-4o, ollama:llama3.2, google_genai:gemini-2.0-flash, litellm:anthropic/claude-sonnet-4-6
Use provider/model format after litellm: (e.g. litellm:huggingface/meta-llama/Llama-3-70b)

Requires OpenCode server running. Start with: opencode --server

API Keys

Keys are encrypted at rest. Only keys needed for your current backend and provider are shown.

Required for Current Backend

Get Key Saved
Required for Claude SDK backend — get a key at console.anthropic.com
Get Key Saved
Get Key Saved
Used for Gemini LLM and image generation. Free at AI Studio.

Tools & Services

Search

Get Key Saved
Get Key Saved
Saved
Used by both web search (parallel provider) and URL extract

Voice & Media

Get Key Saved
Get Key Saved
Indian language TTS (Bulbul), STT (Saaras), OCR (Vision), Translation (Mayura)

Google OAuth

Get Key Saved
Saved
Gmail Calendar Drive Docs

Spotify

Get Key Saved
Saved

Behavior & Safety

Comma-separated list of tools that require approval

Safety

Action:

Memory

LLM used for fact extraction and memory consolidation

Search & Services

Voice

System

Web Server

Set host to 0.0.0.0 to allow access from other devices on your network

Click Restart Server after changing host or port to apply

Budget Controls

Set to 0 for no monthly cap.
Fraction from 0 to 1 (0.8 means warn at 80%).
Use 1-28 to avoid month-length edge cases.
Controls how long trace history is kept.
Cron expression (default: 3 AM daily)
Security Audit

Run 7 security checks on your configuration

/ checks passed
Self-Audit Reports

Health and security check history

No reports yet. Click "Run Now" or enable the daemon.
Report: — / passed
Name advertised in the Agent Card to other agents
Maximum time to wait for an agent response before failing the task
One URL per line. Only these agents can be used for outbound delegation (SSRF protection)

Soul Protocol

Persistent AI identity with psychology-informed memory, OCEAN personality, and emotional state.

e.g. "The Coding Expert", "The Compassionate Creator"
How often to auto-save soul state (0 = disabled, default 300 = 5 minutes)
Import a .soul, .yaml, or .json file. Export saves the current soul to disk.

Screenshot

Screenshot

Files

Reminders

Try: "in 5 minutes", "at 3pm", "tomorrow at 9am"

Intentions

New Intention

Skills

Searching skills.sh...
Popular Skills
Search Results

Skills extend PocketPaw with new /commands. Browse more at skills.sh

Agent Identity

Loading identity...

Identity files shape how the agent thinks and communicates. Changes take effect on the next message.

Memory

Loading memories...

No memories yet

Ask me to "remember" something to save it here

No memories match ""

Entities ()
Relationships ()
No graph entities yet.
Total
Vector Rows
Graph Edges
Backend

Audit Log

No audit entries yet

Security events will be logged here during agent operation

No entries match your filter

Take Paw With You

Scan to Login

Scan with your phone to access PocketPaw remotely via browser.

Cloudflare Tunnel

Telegram Connected

User ID:

You can control PocketPaw from Telegram. Just send a message to your bot!

1

Open @BotFather and send /newbot

2
Enter your bot token above to continue
Scan with Telegram

Scan with your phone camera to open in Telegram

Waiting for connection...

Channels

Saved
Saved
Saved

Get a token from @BotFather. For the full pairing flow, use the tab.

Saved
Saved
Saved
Saved
Usage: curl -X POST <url> -H "X-Webhook-Secret: <secret>" -H "Content-Type: application/json" -d '{"content": "hello"}'
Add ?wait=true to wait for the agent's response.

requires . Install it?

MCP Servers

Beta

System Health

Last checked: (auto-refreshes every 15s while open)

Diagnostics

No issues detected. All checks passed.
Health data not available
Click "Run Check" to diagnose, or wait for the server to reconnect.

Recent Errors

No errors recorded.

Analytics & Traces

Live cost, performance, usage, health, and trace-level diagnostics

Total Spend
Requests
Avg Latency
Error Rate (24h)

Cost by Model

No model usage in this period.

Tool Performance

Top tools
No tool calls in this period.

Usage by Channel

Messages
No channel activity in this period.

System Health Metrics

Recent Errors (1h)
Guardian Block Rate
Memory Store Size
Active Sessions

Cost by Tool

Proportional duration share
No tool cost data in this period.

Channel Uptime (7 days)

No channel uptime data yet.

Monthly Budget

Spent
$0
Reset Day
Auto-Pause

Active Override

Set Temporary Override

No alerts.
No traces for current filters.
Recent traces
Select a trace to inspect details.

Trace Detail

Started
Ended
Duration
Cost

Tool Calls

No tool calls.

LLM Calls

No LLM calls.
🐾

Welcome to PocketPaw Beta

Choose your AI backend to get started.

You're all set!

You can change backends and keys anytime in Settings.