Meet ForkScout
An autonomous AI agent that runs shell commands, browses the web, reads and writes files, remembers everything, and modifies its own code. Not a chatbot wrapper — a real agent.
curl -fsSL https://raw.githubusercontent.com/marsnext/forkscout/main/install.sh | bashBuilt With
The modern stack powering ForkScout
Blazing-fast JS runtime with native TS, built-in bundler, and 3× faster installs.
Strict mode catches bugs at compile time. Every tool and provider is fully typed.
Unified interface for 9+ LLM providers with streaming and multi-step agent loops.
Add capabilities by dropping a JSON config file. Zero code changes required.
Self-hosted meta search. No API keys, no rate limits, fully private web search.
Embedded key-value store for persistent memory. Zero config, survives restarts.
One command to deploy anywhere. Compose stack for agent, search, and memory.
Runtime schema validation for every tool input. Catches bad LLM params early.
Battle-tested ecosystem with native child_process for real shell access.
Next.js incremental bundler with near-instant HMR. Reloads in milliseconds.
Blazing-fast JS runtime with native TS, built-in bundler, and 3× faster installs.
Strict mode catches bugs at compile time. Every tool and provider is fully typed.
Unified interface for 9+ LLM providers with streaming and multi-step agent loops.
Add capabilities by dropping a JSON config file. Zero code changes required.
Self-hosted meta search. No API keys, no rate limits, fully private web search.
Embedded key-value store for persistent memory. Zero config, survives restarts.
One command to deploy anywhere. Compose stack for agent, search, and memory.
Runtime schema validation for every tool input. Catches bad LLM params early.
Battle-tested ecosystem with native child_process for real shell access.
Next.js incremental bundler with near-instant HMR. Reloads in milliseconds.
Built for real autonomy
Every capability is a callable tool — not generated text.
Real Shell Access
Execute any command, install packages, manage servers — not a sandbox.
Persistent Memory
Remembers facts, decisions, and conversations across restarts.
MCP-First
Add capabilities via JSON config. Drop a file, tool is live.
Provider-Agnostic
Swap between 9 LLM providers by changing one JSON field.
Self-Modifying
Edits its own source, typechecks, and restarts itself.
Multi-Channel
Same agent brain on Telegram, terminal, WhatsApp, and web.
Web Search & Browse
Self-hosted SearXNG search and full webpage content extraction.
File Read & Write
Reads files in chunks, writes with directory creation.
Task Orchestration
Parallel workers, sequential chains, live progress — all autonomous.
20 channels. One agent brain.
Same adapter pattern everywhere — raw JSON in, agent reply out. Set env vars and it auto-starts.
Telegram
CoreFull-featured bot with auth, queuing, rate limiting & owner commands
Baileys Web WS protocol, QR pairing, media & image analysis
Terminal
CoreInteractive CLI with live token streaming — same agent brain
Self (HTTP)
CoreREST API for self-sessions, cron jobs, and external triggers
Discord
GatewayGateway WebSocket — guilds, DMs, MESSAGE_CONTENT intent
Slack
Gateway@slack/bolt Socket Mode — no public URL needed
Matrix
Gatewaymatrix-bot-sdk with autojoin — federated, self-hosted chat
WebChat
GatewayBun native WebSocket on /ws — zero external deps
IMAP polling + SMTP reply — works with any email provider
Teams
WebhookMicrosoft Bot Framework — works with Teams & Outlook
Google Chat
WebhookGoogle Workspace API — service account auth
LINE
WebhookLINE Messaging API — group chats & 1:1
Viber
WebhookViber Bot API — direct REST, auto webhook registration
Messenger
WebhookFacebook Messenger Platform — webhook verification built-in
Instagram Graph API — DMs via webhook
SMS
WebhookTwilio Programmable Messaging — reach any phone number
Voice Call
WebhookTwilio Voice — STT input, TTS spoken reply, multi-turn
Twitter / X
PollingX API v2 — DM polling with conversation threading
snoowrap — inbox DMs, mentions & comment replies
YouTube Live
PollingData API v3 — live stream chat with API-set poll interval
Each channel is ~60-80 lines of code. All share the same adapter —createChannelHandler()
What can you build with it?
From DevOps to research — one agent, infinite workflows.
DevOps Automation
Automate deployments, monitor servers, and handle incident response — 24/7 without human intervention.
Code Review & Refactor
Analyze pull requests, suggest improvements, refactor legacy code, and enforce coding standards autonomously.
Research & Analysis
Search the web, read documents, synthesize findings, and deliver structured reports on any topic.
Security Auditing
Scan codebases for vulnerabilities, review dependencies, and generate security compliance reports.
CI/CD Pipelines
Orchestrate build, test, and deploy workflows. Auto-fix failing tests and notify across channels.
Multi-Channel Support
Answer questions on Telegram, WhatsApp, and web simultaneously with shared context and memory.
9 LLM providers. One JSON field.
Swap providers and model tiers without touching code.
// Switch model — no code changes{"llm": {"provider": "openrouter","tier": "balanced"}}
Ready to deploy your own agent?
One command to install. Self-hosted. Full control. No vendor lock-in.