Phantom by AshlrAI

One place for your engineering agents.

Work alongside your agents, or give the fleet an outcome to work toward. Bring your signed-in coding tools, separately billed APIs and local models into one workbench for chats, projects, traces and review.

Start by installing Phantom on your machine, then connect and configure your tools. This page is an introduction; a fleet is ready only after local setup and permissions are in place.

Choose an agent to connectInstall and set up PhantomRead the documentationProtect credentials with Phantom SecretsDownload for Mac

The workbench is distributed as @ashlr/phantom@3.29.15 with the phm CLI. Published release · npm package.The ashlr alias remains compatible; @ashlr/hub is the legacy package. Your installed tools, account setup and permissions determine what can run.

npm install -g @ashlr/phantom@3.29.15

  • SubscriptionsThrough signed-in CLIs
  • API creditsSeparately metered usage
  • Local modelsYour configured runtime
PhantomCoordinate the work

Work with me

Converse, guide and inspect a session.

Work for me

Steer outcomes and follow fleet progress.

Plan Build Verify Review

MCP and CLI tools require configuration.

Phantom SecretsLocal credential boundaryLocusAccount and tenant identityThe ecosystemExplore the connected projects

Ecosystem stars

46 GitHub stars

across 11 public repositories

5 more than Oct 12, 2026.

Oct 12, 2026Nov 2, 2026

Recorded Nov 2, 2026.

Per repository

Star data for agents (JSON)

Your resources.
One extraordinary team.

Bring your subscriptions, API credits and local models into one engineering workbench. Give your agents the tools to turn an idea into a result.

Interactive demo · no agents are running

The illustrative team moves through the task and brings you the result.

AnthropicClaude Code
OpenAICodex
xAIGrok
DevinCloud & CLI
On your machineLocal models
VercelDeployment tools
SupabaseDatabase tools

Connect your supported CLI accounts. Each account keeps its own subscription window and usage reading.

Illustrated handoffs · one shared task

Delegated workflow · a retry fix

Find the cause before changing the code

Scout follows a reported retry failure to a small, reproducible case. The task and its context stay visible while the rest of the team gets ready.

Selected agent
Scout
Task
Investigate the retry failure
Context
Report and reproduction
What you would inspectA reproducible case and a short explanation of the suspected cause.

1 of 5 · Investigate · Ready

The open-source toolkit

Tools with a job. One place to bring them together.

Phantom SecretsKeep secrets out of model context LocusBind tools to the right account LexiconMake technical voice input clearer Explore the ecosystemIndependent projects. Shared purpose.

Provider logos identify resources. MCP servers and ecosystem tools need configuration.

Choose your first agent.

Expand a choice for the path to your first local session. These instructions do not connect an account or install software.

First, install and set up Phantom on your machine. Then follow your agent’s steps below.

Claude Code — connection steps
  1. Install Claude Code and sign in through its vendor CLI on this machine. Keep passwords, credentials and private tokens out of chat.
  2. From a project you authorize Phantom to access, run phm verse. Open the local URL printed by that process; startup may refresh configured accounts.
  3. Open Resources and check that your chosen account or model is ready and its status is fresh. A saved profile alone does not prove a usable connection.
  4. Choose Work with me, create a chat and select a scratch repository. Use Advanced to choose your ready account or model, then ask for a small, checkable change. Review the diff and checks before accepting it.
Continue with Claude Code setup
Codex — connection steps
  1. Install Codex and sign in through its vendor CLI on this machine. Keep passwords, credentials and private tokens out of chat.
  2. From a project you authorize Phantom to access, run phm verse. Open the local URL printed by that process; startup may refresh configured accounts.
  3. Open Resources and check that your chosen account or model is ready and its status is fresh. A saved profile alone does not prove a usable connection.
  4. Choose Work with me, create a chat and select a scratch repository. Use Advanced to choose your ready account or model, then ask for a small, checkable change. Review the diff and checks before accepting it.
Continue with Codex setup
Grok — connection steps
  1. Install Grok and sign in through its vendor CLI on this machine. Keep passwords, credentials and private tokens out of chat.
  2. From a project you authorize Phantom to access, run phm verse. Open the local URL printed by that process; startup may refresh configured accounts.
  3. Open Resources and check that your chosen account or model is ready and its status is fresh. A saved profile alone does not prove a usable connection.
  4. Choose Work with me, create a chat and select a scratch repository. Use Advanced to choose your ready account or model, then ask for a small, checkable change. Review the diff and checks before accepting it.
Continue with Grok setup
API or local model — connection steps
  1. Configure an API engine or a local model runtime in Resources. API usage is billed separately; local models need tool support and available serving capacity. Keep passwords, credentials and private tokens out of chat.
  2. From a project you authorize Phantom to access, run phm verse. Open the local URL printed by that process; startup may refresh configured accounts.
  3. Open Resources and check that your chosen account or model is ready and its status is fresh. A saved profile alone does not prove a usable connection.
  4. Choose Work with me, create a chat and select a scratch repository. Use Advanced to choose your ready account or model, then ask for a small, checkable change. Review the diff and checks before accepting it.
Continue with API or local model setup

Need help? Troubleshoot local setup. Starting a local session does not activate unattended fleet work.

Your engineering tools, in the same conversation.

Bring configured MCP servers and CLIs for source control, deployment, databases, design and observability. The workbench coordinates tool access; each server, account and permission determines the actions available.

  • Vercel
  • Cloudflare
  • Supabase
  • Stripe
  • Slack
  • Discord
  • PostHog
  • Sentry
  • GitHub
  • Docker
  • Notion
  • Linear
  • Figma

Logos identify products, not endorsement or automatic connection. Subscription CLI sessions, API credentials, MCP tools and local-model runtimes remain separate resources.

Explore the engineering toolkit

Phantom Secrets. A boundary for your credentials.

The local-first Secrets CLI and MCP server keep real API keys out of supported agent paths. See its installation, client setup and reviewed release evidence on the dedicated product page.

Explore Phantom Secrets →Agent quickstart

Looking for an existing Secrets section?

How it worksSecrets featuresInstall SecretsConnect a clientCompare the credential boundarySecrets pricingSecrets questions