Skip to main content

Web-based agent session visualizer for Claude Code

Project description

AgentsTV

PyPI version License: MIT Python 3.10+

Turn your AI coding agent sessions into a Twitch-style pixel art live stream.

Your Claude Code sessions are already generating thousands of events. AgentsTV turns them into something you can actually watch -- pixel art characters coding in real-time, simulated Twitch chat reacting to what the agent does, esports-style narration, and a Master Control Room when you're running multiple sessions.

AgentsTV session view

AgentsTV dashboard

Quick Start

pip install agentstv
agentstv

That's it. The dashboard auto-discovers your Claude Code session logs and opens in your browser.

Add AI Chat (optional)

AgentsTV can use a local LLM via Ollama for viewer chat, narrator commentary, and interactive Q&A:

ollama pull qwen3:14b
agentstv --llm ollama --ollama-model qwen3:14b

Without Ollama, everything works -- viewer chat uses built-in fallback messages.

Hardware Model Notes
Desktop GPU (12GB+ VRAM) qwen3:14b Best overall
MacBook / Laptop qwen3:8b Good balance of quality and speed
Low VRAM / older GPU phi4-mini Lightweight, surprisingly capable

Also supports OpenAI and Anthropic as cloud LLM providers. Configure via CLI flags or the in-app settings panel.

Features

  • Pixel art webcam -- Procedurally generated coding scenes with idle animations, desk decorations, weather effects, and event reactions (error shakes, completion confetti, spawn rings, thinking bubbles)
  • Viewer chat -- Simulated Twitch chat where LLM-generated viewers react to what the agent is doing
  • Interactive chat -- Ask questions about what the agent is working on and get context-aware answers
  • Narrator bot -- Esports-style play-by-play commentary on agent activity
  • Auto stream titles -- Generated from recent activity (e.g. Coding / Python / main)
  • Master Control Room -- Multi-monitor view aggregating all active sessions with status LEDs and alert lights
  • Session replay -- Play back completed sessions with seek, speed control (0.5x-8x), and full event recreation
  • Clip recording -- Record and download WebM clips of interesting moments
  • OBS overlay -- Browser Source page at /overlay.html for compositing over your stream
  • Sound effects -- Keystrokes, error buzzer, completion chime, chat pops
  • Public mode -- --public flag scrubs API keys, tokens, paths, and secrets server-side before sharing
  • Dark / light theme, keyboard shortcuts (? to view), search and sort, stream alert toasts

Usage

agentstv                         # Launch (opens browser)
agentstv --port 8420             # Custom port (default: 8420)
agentstv --host 0.0.0.0         # Bind to all interfaces (LAN access)
agentstv --public                # Redact secrets for public sharing
agentstv --low-power             # Reduce LLM batch sizes for laptops
agentstv --no-browser            # Don't auto-open browser

Session logs are discovered from ~/.claude/projects/. Override with AGENTSTV_DATA_DIR=/path/to/logs.

Remote LLM Setup

You can run AgentsTV on the machine where your sessions live and offload LLM inference to a separate machine (e.g. a MacBook running Ollama). No file syncing required -- only lightweight API calls go over the network.

On the LLM machine (e.g. MacBook with Ollama):

# Allow remote connections
OLLAMA_HOST=0.0.0.0 ollama serve

On the Claude Code machine (where your sessions are):

# First time — pass the IP of your LLM machine
agentstv --remote <llm-machine-ip> --ollama-model qwen3:8b

# After that — remembers the remote config
agentstv --remote

Then open http://<claude-code-machine-ip>:8420 from any browser on your network.

This is useful when your coding machine has no GPU or you want to keep LLM inference on a power-efficient device like an Apple Silicon Mac. The --remote flag sets --host 0.0.0.0 and saves the remote Ollama URL to ~/.agentstv/remote.json.

Supported Formats

  • Claude Code JSONL transcripts (full support)
  • Codex CLI session logs (experimental)
  • Gemini CLI session logs (experimental)

About

The idea started as a simple TUI to replay Claude Code session logs. I built the first version, went to bed, and woke up wanting something completely different: a Twitch-style live dashboard with pixel art characters, LLM-powered chat, and streaming overlays. The full web app was built and published in about 12 hours.

Built with Claude Code. I made the architectural decisions and directed the implementation across Python, JavaScript, HTML, and CSS.

Contributing

Contributions are welcome! AgentsTV is MIT-licensed and open to PRs.

git clone https://github.com/andrewle8/AgentsTV.git
cd AgentsTV
pip install -e .
agentstv --no-browser

Areas where help is especially welcome:

  • Additional agent format parsers (Copilot CLI, Aider, etc.)
  • Custom pixel art themes and character skins
  • Mobile layout improvements
  • OBS overlay enhancements

License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

agentstv-0.7.3.tar.gz (531.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

agentstv-0.7.3-py3-none-any.whl (102.9 kB view details)

Uploaded Python 3

File details

Details for the file agentstv-0.7.3.tar.gz.

File metadata

  • Download URL: agentstv-0.7.3.tar.gz
  • Upload date:
  • Size: 531.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.10

File hashes

Hashes for agentstv-0.7.3.tar.gz
Algorithm Hash digest
SHA256 ae604a0272c22d59b54c608718a55b3911bc4db9f3622e3f337c8841858359b9
MD5 a2441b61dbd6b7035d7b459ac4b97f63
BLAKE2b-256 483293504679c160de6092b1bfc26cd790783d1d61ff68dd9f499d83dc320f84

See more details on using hashes here.

File details

Details for the file agentstv-0.7.3-py3-none-any.whl.

File metadata

  • Download URL: agentstv-0.7.3-py3-none-any.whl
  • Upload date:
  • Size: 102.9 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.10

File hashes

Hashes for agentstv-0.7.3-py3-none-any.whl
Algorithm Hash digest
SHA256 d4abf25b17d73032271f96031c93e25d04a55ade05e7bb6a9f5eb12827e0b357
MD5 f364ccd5e59bf654e5ce0783a45ca6ed
BLAKE2b-256 efa5abc40f48a4084eda276b487b6542177d5bcb781de84a0ecdc58d9e3c21f6

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page