phantom

---

The Idea

AI agents today are disposable. You open a chat, get an answer, close the tab, and the context is gone. Next time you start from scratch. Every session is day one.

Phantom takes a different approach: give the AI its own computer. A dedicated machine where it installs software, spins up databases, builds dashboards, remembers what you told it last week, and gets measurably better at your job every day. Your laptop stays yours. The agent's workspace is its own.

This is not a chatbot. It is a co-worker that runs on Slack, has a web chat interface at /chat, has its own email address, creates its own tools, and builds infrastructure without asking for permission. Don't take our word for it - scroll down to see what production Phantoms have actually built.

What This Actually Looks Like

These are not mockups. They happened on production Phantom instances.

Built an analytics platform from scratch

A Phantom was asked to help with data analysis. It installed ClickHouse on its own VM, downloaded the full Hacker News dataset, loaded 28.7 million rows spanning 2007-2021, built an analytics dashboard with interactive charts, and created a REST API to query the data. Then it registered that API as an MCP tool so it could use it in future sessions and other agents could query it too.

Nobody asked it to build any of this. It identified analytics as useful and built the entire stack.

28.7 million items. 755K unique authors. 4.3 million stories. Built, loaded, and served by a Phantom on its own machine.

Extended itself with a channel it was never built with

Phantom ships with Slack, Telegram, Email, and Webhook channels. It does not ship with Discord. When asked "Can I talk to you on Discord?", the Phantom said: "Not right now. Discord isn't wired up. That said, I could build it."

It explained the Discord Bot API, walked the user through creating a Discord application, provided a magic link for secure token submission, and said: "Once you save it, I'll automatically spin up the container and you'll be live on Discord."

After submitting the token, Phantom went live on Discord. It permanently gained a communication channel it was never designed with.

The agent was honest about what it could not do, then built the capability on the spot.

Started monitoring its own infrastructure

A Phantom discovered Vigil, a lightweight open-source system monitor with 3 GitHub stars. It understood what Vigil does, integrated it into its existing ClickHouse instance, built a sync pipeline that batch-transfers metrics every 30 seconds, and created a real-time monitoring dashboard showing service health, Docker container status, network I/O, disk I/O, system load, and data pipeline health.

890,450 rows. 25 metrics. Auto-refreshing. The agent is watching its own infrastructure.

It found a 3-star open-source project, integrated it into its data pipeline, and built observability for itself.

---

This is what happens when you give an AI its own computer.

Bring Your Own Model

Phantom is not locked to any single AI backend. It ships with support for seven providers out of the box, configured through a single YAML block:

• Anthropic (default) - Claude Opus, Sonnet, Haiku
• Z.AI - GLM-5.1 and GLM-4.5-Air via Z.AI's Anthropic-compatible API. Roughly 15x cheaper than Claude Opus for comparable coding quality.
• OpenRouter - 100+ models through one key
• Ollama - Any GGUF model on your own GPU, zero API cost
• vLLM - Self-hosted inference with OpenAI-compatible endpoints
• LiteLLM - Local proxy bridging OpenAI, Gemini, and more
• Custom - Any Anthropic Messages API compatible endpoint

Switching providers is two lines of YAML:

# phantom.yaml
model: claude-opus-4-7
provider:
  type: zai
  api_key_env: ZAI_API_KEY
  model_mappings:
    sonnet: glm-5.1

Set ZAI_API_KEY in .env, restart, done. Both the main agent and every evolution judge flow through the chosen provider from that point on. The tools are the same, the memory is the same, the self-evolution pipeline is the same. Only the brain changes.

Anthropic stays the default. Existing deployments continue to work with no configuration changes. See docs/providers.md for the full reference.

Quick Start

Docker (recommended)

curl -fsSL https://raw.githubusercontent.com/ghostwright/phantom/main/docker-compose.user.yaml -o docker-compose.yaml
curl -fsSL https://raw.githubusercontent.com/ghostwright/phantom/main/.env.example -o .env
# Edit .env - add your ANTHROPIC_API_KEY, Slack tokens, and OWNER_SLACK_USER_ID
docker compose up -d

Your Phantom is running. Qdrant starts for memory, Ollama pulls the embedding model, and the agent boots. Check health at http://localhost:3100/health. With Slack configured, it DMs you when it's ready. Add RESEND_API_KEY for email sending. See Getting Started for full setup.

Security note, Docker socket mount: docker-compose.yaml mounts /var/run/docker.sock into the Phantom container so it can spawn sibling containers (e.g. sandboxed code execution). This is an intentional architectural trade-off: the socket grants the container root-equivalent access to the Docker daemon, which means a compromised Phantom process could create, modify, or destroy any container on the host. Mitigations: run Phantom on a dedicated machine or VM (not your personal workstation), and do not expose the host's Docker socket to untrusted workloads. See docs/security.md for the full threat model.

Managed (free)

Get a Phantom on a dedicated VM with nothing to install. Bring your Anthropic API key, we give you the machine.

ghostwright.dev/phantom

What People Build with Phantom

Phantom is not just for engineers. It is for anyone who wants an AI that remembers, learns, and builds things you can actually share.

The key difference: when an AI runs on your laptop, everything it builds is trapped on localhost. Only you can see it. Phantom runs on a VM with a public domain. Dashboards, tools, pages, APIs - they all get a URL you can send to your team, your manager, or your clients. Your laptop is not a server. Phantom's VM is.

For People Who Don't Write Code

You don't need to install developer tools, learn a build system, or figure out hosting. You describe what you want in Slack. Phantom builds it, deploys it on its own machine, and gives you a link.

• "Build me a Chrome extension that highlights overdue emails." Phantom builds it and gives you a zip file. You drag it into Chrome. No Xcode, no npm, no terminal. Just install it and go.
• "Make a landing page for my side project." Phantom builds the page, serves it on its public domain, and gives you a URL. Send it to anyone. No hosting setup, no domain configuration.
• "Create a weekly report of our open support tickets." Phantom builds the automation, runs it on a schedule, and emails you the summary every Friday from its own email address.
• "Build a form where the team can submit feature requests." Phantom creates it on its domain, handles submissions, and routes them to Slack.

Software Engineering

• Morning standup: "Every weekday at 9am, summarize open PRs, CI status, and what needs review." Phantom checks GitHub, compiles the summary, and posts it to your team channel.
• Codebase onboarding: "Clone this repo and give me an architecture overview." Phantom reads the code and returns specifics, not generics. "Next.js 16 with App Router, Drizzle ORM on Neon Postgres, flaky step in the migration check."
• Infrastructure: "Set up a dev Postgres for this project." Phantom spins up a Docker container on its own machine, creates the schema, runs migrations, and gives you a connection string. Nothing installed on your laptop.
• Data pipelines: "Pull from our API every hour and load into Postgres." Phantom builds the pipeline, runs the database on its VM, and schedules the cron. You get a connection string.
• Custom MCP tools: "Create an MCP tool that queries our internal API." Phantom builds the tool, registers it, and any Claude Code instance connected to it can call it immediately.

Sales and Account Management

• Prospect research: "Research these 10 companies and build outreach strategies." Phantom gathers intelligence, identifies decision-makers, and creates briefing pages you can share with your team.
• Recruiting profiles: "Make a profile page for this candidate." Phantom researches the person and builds a public profile page you can send to hiring managers. Shareable URL with auth.
• Competitor monitoring: "Every 2 minutes, check for open-source repos similar to our product." Phantom runs a cron job, tracks new entrants, and posts updates to Slack.

Data and Analytics

• Build your own analytics stack: Phantom installs databases, builds ETL pipelines, and registers APIs as MCP tools for future sessions. All on its own machine. (See the ClickHouse story above.)
• Shareable dashboards: "Track our PR velocity and show it to the team." Phantom builds an ECharts dashboard, serves it on a public URL with auth, and sends you the link. Your team bookmarks it. They see the same thing you see.
• Data exploration: "Load this dataset and let me ask questions about it." Phantom creates a queryable environment on its VM. You ask in plain English, it translates to SQL.

Operations, Marketing, and Everyone Else

• Automated reports: "Email a weekly summary of open issues to team leads every Friday." Phantom compiles the report and sends it from its own email address.
• Competitor watching: "Monitor these five websites for changes and tell me when something updates." Phantom checks on a schedule and notifies you in Slack.
• Blog drafts: "Write a post based on our recent product updates." Phantom reads the changelog, writes the draft, and serves it as a page you can review and share with your editor.
• Personal coaching: "Create a training plan for me." Phantom builds a complete plan with a coach persona, served as a page you can reference daily.

For Everyone

• Remembers what you told it last Monday and uses it on Wednesday.
• Never asks the same question twice.
• Gets measurably better at YOUR job every day, not just at being a generic assistant.

How It's Different

Most AI assistants