Parchi

DO NOT SKIP THIS WARNING, DO NOT USE THIS SOFTWARE IF YOU DO NOT READ THIS:

BROWSER AUTOMATION CAN PUT YOUR SOCIAL MEDIA ACCOUNTS AT RISK
BROWSER AUTOMATION CAN CAN BE AGAINST TERMS OF SERVICE OF MANY WEBSITES
THIS CAN LEAD TO PROMPT INJECTION ATTACKS
THIS CAN LEAD TO LEAKING YOUR PERSONAL INFORMATION
THIS CAN CAUSE TECHNICAL ISSUES WITH YOUR BROWSER

THIS IS ALL REAL YOU MUST UNDERSTAND THE RISKS OF USING BROWSER AUTOMATION AND TRUSTING AN LLM WITH YOUR BROWSER.

THIS TOOL WAS BUILT BY AND FOR PEOPLE WHO UNDERSTAND AND ACCEPT THESE RISKS.

Parchi

Your AI-powered browser copilot.

Chat-driven browser automation that lives in your sidebar. Navigate, read, click, extract — all through natural language.

Installation · Setup · Features · Docs · Relay CLI

📦 Installation

Chrome (recommended)

git clone https://github.com/0xSero/parchi.git
cd parchi
npm install
npm run build

Then:

open chrome://extensions
enable Developer mode
click Load unpacked
select dist/
pin and open Parchi from the toolbar

Firefox

npm run build:firefox

Open about:debugging#/runtime/this-firefox
Click "Load Temporary Add-on"
Select any file inside the dist/ folder

XPI packaging: npm run build:firefox:xpi outputs dist-firefox/<extension-name>-<version>.xpi for distribution. Requires Developer Edition/Nightly or Mozilla add-on signing for release installs.

🔧 Setup Your AI Provider

Parchi works with any OpenAI-compatible endpoint. Open the Settings panel (gear icon) to configure.

Option A: Direct API Key (BYOK)

Use your own API key from any supported provider:

Provider	API URL	Example Models
OpenAI	`https://api.openai.com/v1`	`gpt-5.4`, `gpt-5-nano`, `gpt-5.3-codex`
Anthropic	`https://api.anthropic.com/v1`	`claude-sonnet-4-6`, `claude-opus-4-6`
Kimi	`https://api.moonshot.cn/v1`	`kimi-for-coding` (recommended)
OpenRouter	`https://openrouter.ai/api/v1`	Any model on OpenRouter
Local (Ollama, LM Studio)	`http://localhost:11434/v1`	`qwen`, `mistral`, etc.
Any OpenAI-compatible	Your endpoint URL	Your model name

Steps:

Open Settings → select your provider or choose Custom
Paste your API Key
Set the API URL (auto-filled for known providers)
Pick a Model from the dropdown (auto-fetched) or type one manually
Hit Save — start chatting

Option B: Parchi Account (Proxy Mode)

Sign in with Google or GitHub to use Parchi's hosted proxy. No API keys needed — billing is handled through your Parchi subscription.

Open Settings → choose Paid account mode
Sign in via OAuth
Select a model and start chatting

🔑 Use Your Existing AI Subscription

Already paying for Claude Pro, ChatGPT Plus, Gemini Advanced, or another AI subscription? You can route that subscription through Parchi using CLIProxyAPI.

CLIProxyAPI is an open-source proxy that converts your existing AI subscription access into an OpenAI-compatible API endpoint — no separate API key purchase needed.

How it works

Install CLIProxyAPI — follow the setup guide
Authenticate with your existing provider (Claude, OpenAI, Gemini, etc.) via OAuth
Point Parchi at the proxy endpoint:
- Open Settings → provider: Custom
- API URL: your CLIProxyAPI endpoint (e.g. http://localhost:PORT/v1)
- API Key: your proxy token
- Model: the model you want to use
Chat — requests route through your existing subscription

Supported subscriptions

Subscription	What you get
Claude Pro / Max	Claude models via your Anthropic account
ChatGPT Plus / Pro	GPT-4o, o1, etc. via your OpenAI account
Gemini Advanced	Gemini models via your Google account
Qwen / iFlow	Additional provider access

CLIProxyAPI supports multi-account round-robin, streaming, function calling, and multimodal inputs. See their documentation for full details.

✨ Features

Chat + AI — streaming answers, reasoning display, profiles, vision support, context compaction, workflow shortcuts
Browser tools — navigate, click, clickAt, type, scroll, screenshot, findHtml, tab management, video helpers, planning, subagents
Session tools — session tabs, floating HUD, saved history, markdown export with tool traces
Controls — tool permissions, allowlists, confirmations, themes, zoom, custom headers

See docs/agent-pipeline.md for the runtime shape and docs/tab-process-performance-playbook.md for perf triage.

🛰 Relay Daemon & CLI

Expose Parchi as a local automation endpoint for scripts and external tools.

# 1. Build with relay token
PARCHI_RELAY_TOKEN=your-secret npm run build

# 2. Start the daemon
PARCHI_RELAY_TOKEN=your-secret npm run relay:daemon

# 3. Open the extension once — it can auto-pair from localhost /v1/pair.
#    (Manual fallback in Settings → Relay:
#      URL: http://127.0.0.1:17373
#      Token: your-secret)

Safer managed daemon helper (recommended):

npm run relay:secure -- start   # generates/stores strong token, loopback-only
npm run relay:secure -- status
npm run relay:secure -- rotate  # rotates token (+ restarts if running)
npm run relay:secure -- stop

CLI commands:

export PARCHI_RELAY_TOKEN=your-secret

# List connected agents
npm run relay -- agents

# List available tools
npm run relay -- tools

# Execute a single tool
npm run relay -- tool navigate --args='{"url":"https://example.com"}'

# Run the agent and wait for result
npm run relay -- run "Open example.com and summarize the page"

Electron desktop automation (relay-native + direct CLI)

You can now control Electron apps (Slack, VS Code, Discord, etc.) in two modes:

Relay-native agent (multi-agent routing through parchi-relay)
Direct CLI mode (parchi electron ... passthrough to agent-browser)

# Build binaries
npm run build

# Start relay daemon (terminal A)
PARCHI_RELAY_TOKEN=your-secret npm run relay:daemon

# Start Electron relay agent (terminal B)
PARCHI_RELAY_TOKEN=your-secret npm run electron:agent

# Confirm agents and choose Electron as default when needed
npm run relay -- agents
npm run relay -- default-agent set <electron-agent-id>
npm run relay -- tools
npm run relay -- tool electron.launch --args='{\"app\":\"Slack\",\"port\":9222}'
npm run relay -- tool electron.connect --args='{\"cdpEndpoint\":\"9222\"}'
npm run relay -- tool electron.snapshot --args='{\"interactive\":true}'

Managed mode (recommended once relay secure is configured):

# Reads relay token/url from ~/.parchi/relay-secure.json
npm run electron:secure -- start
npm run electron:secure -- status
npm run electron:secure -- stop

Direct mode:

# Launch app with remote debugging
parchi electron launch "Slack" --port=9222

# Pass any agent-browser command through parchi
parchi electron connect 9222
parchi electron snapshot -i
parchi electron click @e5

🏗 Architecture

High level:

Sidepanel UI renders chat, history, settings, and tool timelines
Background worker runs the agent loop, tool execution, relay handling, and stream updates
Shared contracts define plans, prompts, tools, and runtime-message schemas

Workspaces:

Workspace	Role
`packages/backend/`	Convex backend (auth, billing, API proxy)
`packages/cli/`	Local CLI entrypoint and daemon client
`packages/electron-agent/`	Relay-native Electron desktop automation agent
`packages/extension/`	Browser extension runtime, UI, and tools
`packages/relay-service/`	Relay daemon + relay protocol CLI
`packages/shared/`	Shared plans, prompts, schemas, and message types
`packages/website/`	Static website + billing pages

More detail:

🔨 Development

npm install                   # install all workspace deps
npm run build                 # build extension + relay + CLI bundles
npm run typecheck             # repo-wide type checking
npm run lint                  # biome linter
npm run lint:fix              # auto-fix lint issues
npm run test:unit             # run unit tests
npm run test:integration      # run headless integration tests
npm run test:e2e              # run browser E2E harness
npm run check:repo-standards  # enforce changed-file guardrails
npm run perf:tabs             # sample Firefox/Chrome tab CPU + RAM
npm run backend:dev           # run Convex dev backend workspace
npm run dev -w @parchi/website  # run website workspace locally

After building, reload the extension in chrome://extensions to pick up changes.

Firefox builds

npm run build:firefox       # build for Firefox → dist/
npm run build:firefox:xpi   # package as .xpi for distribution

Performance leak audits

TAB_AUDIT_SAMPLES=6 TAB_AUDIT_INTERVAL_MS=10000 npm run perf:tabs

The audit now includes Parchi-attributed Firefox totals and RSS/CPU slope so you can separate overall browser pressure from likely extension pressure.

Use the full workflow in docs/tab-process-performance-playbook.md to run active/idle audits and validate regressions in Firefox + Chrome.

📊 Chrome vs Firefox

Feature	Chrome	Firefox
Tab grouping	✅	—
Side panel	✅ Native	Sidebar (adapted)
Relay keepalive	✅ Offscreen doc	—
Min version	MV3	109.0+

Parchi is MIT licensed.

Built with the AI SDK · Styled with Warm Paper design system

Name		Name	Last commit message	Last commit date
Latest commit History 346 Commits
.factory		.factory
.github		.github
.husky		.husky
design-system		design-system
docs		docs
packages		packages
scripts		scripts
temp_pi_agent_core/package		temp_pi_agent_core/package
temp_pi_ai/package		temp_pi_ai/package
tests		tests
types		types
.c8rc.json		.c8rc.json
.env.publish.example		.env.publish.example
.gitignore		.gitignore
1		1
AGENTS.md		AGENTS.md
LICENSE		LICENSE
PRIVACY.md		PRIVACY.md
README.md		README.md
SKILL.MD		SKILL.MD
amo-metadata.json		amo-metadata.json
biome.json		biome.json
knip.json		knip.json
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Parchi

📦 Installation

Chrome (recommended)

Firefox

🔧 Setup Your AI Provider

Option A: Direct API Key (BYOK)

Option B: Parchi Account (Proxy Mode)

🔑 Use Your Existing AI Subscription

How it works

Supported subscriptions

✨ Features

🛰 Relay Daemon & CLI

Electron desktop automation (relay-native + direct CLI)

🏗 Architecture

🔨 Development

Firefox builds

Performance leak audits

📊 Chrome vs Firefox

About

Uh oh!

Releases 8

Packages

Uh oh!

Contributors 3

Languages

Folders and files

Latest commit

History

Repository files navigation

Parchi

📦 Installation

Chrome (recommended)

Firefox

🔧 Setup Your AI Provider

Option A: Direct API Key (BYOK)

Option B: Parchi Account (Proxy Mode)

🔑 Use Your Existing AI Subscription

How it works

Supported subscriptions

✨ Features

🛰 Relay Daemon & CLI

Electron desktop automation (relay-native + direct CLI)

🏗 Architecture

🔨 Development

Firefox builds

Performance leak audits

📊 Chrome vs Firefox

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 8

Packages 0

Uh oh!

Contributors 3

Languages

Packages