Desktop access
Screen, browser, VNC, and full desktop tools let agents operate real UI workflows when APIs are not enough.
Open source. Self-hosted. Built for real work.
Self-hosted agent runtime for real infrastructure.
Run durable AI agents against your own files, tools, browsers, desktops, services, and workflows without moving the runtime into a vendor-controlled agent cloud.
install the latest release
Loading current release command...
Relay-backed tools
Shared context
Multi-provider agents
Deterministic flows
First-class Telegram client
Why PawFlow
Agents should work where your infrastructure already is.
PawFlow keeps the orchestration layer under your control. The server coordinates agents and conversations; relays execute tools next to the workspace; flows run repeatable work with explicit structure.
Your files stay reachable through relays.Agents can read, edit, search, run commands, use browser/desktop actions, and generate media through controlled tool routes.
Context survives restarts.Conversations, memory, knowledge graphs, diaries, project graphs, and FileStore outputs remain available across clients.
Providers are interchangeable.Use Codex app-server, Claude Code interactive, Antigravity/Agy, Gemini CLI, Anthropic, OpenAI, or compatible endpoints per agent or conversation.
Architecture
A clear boundary between server, relay, agent, and flow.
PawFlow does not need permanent direct access to every machine. It routes work through connected relays and explicit services.
Clients
Web, CLI, VS Code, API
Web, CLI, VS Code, API
PawFlow Server
agents, context, flows, auth
agents, context, flows, auth
Relay
filesystem, shell, screen, browser
filesystem, shell, screen, browser
Workspace
code, desktop, local services
code, desktop, local services
Design with agents. Execute with structure.
Use agents for exploration, coding, decisions, and maintenance. When work becomes repeatable, turn it into a flow: CRON triggers, task DAGs, backpressure, checkpoints, retries, and explicit LLM calls only where modeled.
View the daily digest pattern
Webchat workspace
The conversation can open the same surfaces the agent uses.
A linked relay is not only file access. From the webchat, operators can open desktop sessions, terminals, VS Code/code-server, context editors, memory editors, and provider runtime views while the agent works.
Desktop RelayAgents can inspect and control a full desktop through screen tools. With a local relay started with `allow_local`, that desktop can be the local machine where the relay runs; otherwise it is the relay container or configured remote desktop.
Terminals on the relayThe webchat can open terminals for the relay Docker workspace, the host-local relay surface when allowed, and the server relay shell for working directly on the PawFlow server.
Interactive provider tmuxCLI-backed interactive providers such as Claude Code interactive and Antigravity/Agy run in tmux-backed sessions, so their live terminal state can be observed and debugged instead of being a black box.
What you can do
From first chat to operational workflows.
Agentic coding
Run coding agents against a linked workspace with relay-backed read, edit, shell, grep, browser, and project graph tools.
Multi-agent work
Delegate tasks, assign plans, verify outputs, and run different providers in the same conversation.
Vision for every model
Most agent stacks silently drop images sent to a text-only model. PawFlow delegates them to a separate vision LLM that returns text, layout, states, and UI pixel coordinates — so any reasoning model, including free-tier ones, can see uploads and operate a desktop.
Desktop and browser tasks
Operate controlled desktops, browsers, VNC sessions, and forwarded local services from the same runtime.
Media generation
Generate and transform images, video, audio, voice, 3D assets, try-on outputs, and upscaled media into FileStore.
Deterministic automations
Run NiFi-style JSON flows with 100+ task types across IO, data, control, system, and AI categories.
Self-hosted security model
Use explicit relays, stored secrets, provider boundaries, private gateways, and per-agent permissions.
First-class Telegram
Run your agents from a Telegram chat.
Telegram is a complete, supported client for the same agent runtime as web chat and PawCode — not a webhook bolt-on. Message a bot and reach the same durable conversations, agents, and tools, with live thinking and tool events streamed back into the chat.
Same runtimeShared conversations, multi-agent context, attachments, and FileStore outputs.
Rich messagesText, documents, photos, voice, slash commands, and spoken replies.
Voice in, voice out
Talk to your agents — and let them talk back.
Hold a live spoken conversation with an agent: continuous speech-to-speech over OpenAI Realtime or Gemini Live, with live captions, barge-in, spoken tool use, and transcripts saved as normal conversation history — voice-native agents even answer Telegram voice notes in their own voice. Classic speech-to-text and text-to-speech remain first-class too: dictate from the webchat microphone, get replies read aloud, run fully self-hosted voices (Supertonic, Voicebox, LuxTTS) or any OpenAI-compatible endpoint.
Realtime conversationFull-duplex voice sessions with barge-in, push-to-talk, tools, and Telegram voice-note replies.
Speech-to-textWebchat microphone and Telegram voice notes, transcribed by your STT service.
Text-to-speechSpoken replies in webchat and Telegram, plus voice tools for agents.
Built-in cloud desktop
A real self-hosted cloud desktop, one click from the chat.
Every desktop-capable relay — including the server relay built into PawFlow itself — ships a full Linux virtual desktop. Open it as a noVNC tab from the webchat workspace menu, stream its audio, and keep working: no extra VM, no third-party remote-desktop service, everything stays on your infrastructure. Text-only agents can delegate screenshots to a separate vision LLM, then reason over its UI description and coordinates.
For younoVNC desktop and audio streaming in a browser tab, next to the conversation.
For agentsThe same desktop through
screen and see tools, with approved clicks, typing, and shell.
Private Gateway
Your server stays invisible until you sign in.
Internet-facing routes sit behind an explicit gateway step with rate limits, cooldowns, and bans. Sign in with Google, GitHub, X.com, Telegram, Microsoft, Facebook, Amazon, or built-in PawFlow accounts — and disguise the gateway itself with themed skins (Matrix, Netflix, Blade Runner, terminal, wifi portal…) so the login page looks like anything but an agent platform.
Auth providersGoogle, GitHub, X.com, Telegram, Microsoft, Facebook, Amazon, or built-in accounts.
Gateway skinsCamouflage themes for the public-facing page — a UX layer on top of the security model, not a substitute for it.
Make it yours
Theme the whole workspace — globally or per conversation.
Pick a theme for the entire webchat or give each conversation its own look. Themes are plain resources — a theme.json plus CSS and assets — so you can create, import, and share them through PFP packages and resource depots without touching product code.
Global themeOne look for the whole workspace, applied from the chat theme menu.
Per conversationEach conversation can override the global theme — brand a client project, dim a long-running ops channel.
Built-in IDE
A full VS Code IDE, served by your own relay.
Every relay workspace ships code-server: type /code or pick VS Code from the webchat workspace menu and a complete browser IDE opens on the exact files your agents are working on. Review diffs, search, and edit by hand — no local install, nothing leaves your infrastructure.
You reviewFull editor, search, and git view on the relay workspace, in a browser tab.
Agents keep workingManual edits and agent tools share the same workspace boundary and conversation.
VS Code extension
Your agents, in the editor you already use.
Install the release .vsix and PawFlow becomes a panel inside your own VS Code: the same durable conversations, agents, files, and tools as web chat — with live thinking, streaming responses, and tool events right next to your code.
Same backendOne conversation shared with web chat, PawCode, and Telegram — not a separate agent runtime.
In contextSelection actions and the chat sidebar work over your editor, the relay workspace, and your tools.
PawCode CLI
The same agents, straight from your terminal.
PawCode is a Claude Code-style terminal client for the shared runtime: login once, resume the same durable conversations, and watch thinking, tool calls, and streaming responses render live in the shell — no browser required.
Shared conversationsPick up the exact conversation you left in web chat, VS Code, or Telegram.
Server & gateway awarePoint it at any server, including Private Gateway-protected ones, and it remembers the settings.
The practical difference
Not every workflow should keep an agent in the loop forever.
For recurring jobs, PawFlow lets the agent design or maintain the automation, then the flow engine runs the explicit graph. You decide where LLM calls are allowed and where execution must stay deterministic.
Agent modeExplore, code, inspect, reason, delegate.
Flow modeSchedule, route, retry, transform, deliver.
Five-minute path
Install it, open the wizard, start the first conversation.
The Docker installer checks prerequisites, starts PawFlow, opens the bootstrap wizard, creates the admin user, configures the first LLM service, and deploys the starter agent flow.