Complete Mac Guide: AntStation Desktop + Hermes via SSH Reverse Tunnel

Your self-hosted AI agent, powered by peer-to-peer inference — zero server-side install, full P2P network access from your MacBook.

April 2026 • Tutorial • 8 min read

🎯 My Exact Setup

This setup gives my VPS-hosted agent access to the entire AntSeed P2P network — without installing anything on the server.

What is AntStation?

AntStation is the desktop client for the AntSeed peer-to-peer AI services network. Instead of relying on a single AI provider (Anthropic, OpenAI, etc.), AntStation connects you directly to AI providers on a decentralized network.

Why this matters:


Part 1: AntStation Desktop App (Mac Setup)

Installation

  1. Download AntStation from antseed.com
  2. Open the .dmg file
  3. Drag to Applications
  4. Launch AntStation

First Launch

When you open AntStation for the first time, you'll see a welcome screen:

Welcome to AntStation
AntSeed is a peer-to-peer network for AI services.

[Connect Wallet]    [Skip for Now]

Connect Wallet links your crypto wallet (Rabby, MetaMask, or Coinbase Wallet) to manage deposits and payments.

Skip for Now lets you explore first — you can add funds later.

The AntStation Interface

SectionPurpose
ChatDirect chat with your pinned peer (like this session with Surplus Intelligence)
NetworkBrowse all available peers and their models, pricing, reputation
PaymentsManage your USDC deposits and see your balance
SettingsNode identity, preferences, proxy port

Part 2: Depositing USDC (Fund Your Account)

New users get $1 trial credit, but for regular use you'll need to deposit USDC.

How to Deposit

  1. Open AntStation
  2. Click the Payments tab
  3. Click Add Funds or Deposit USDC

A local web interface opens at http://localhost:3118:

  1. Connect your wallet (Rabby, MetaMask, or Coinbase Wallet)
  2. Enter the amount to deposit (e.g., $10 USDC)
  3. Confirm the transaction

💡 Supported Wallets

Understanding Your Balance

StatusMeaning
AvailableUSDC you can spend right now
ReservedUSDC locked for active sessions (up to your credit limit)
Credit LimitMax you can have reserved at once (starts at $10, grows with usage)

The credit limit grows over time: +$5 per unique seller, +$0.50 per day of history, up to $50 max. This is how AntStation handles payment risk in a trustless P2P environment.


Part 3: Browsing & Picking Peers

Using the Network Tab

Click the Network tab in AntStation to see all available peers. You'll see something like this:

Peer ID Name Providers Services ───────────────────────────────────────────────────────────────────────────────────── 4668854ba3e8b094e6f48fbeb59c... Dark Signal openai-responses, openai gpt-5.4, gpt-5.5 +3 1d90f467689d499dc435e5744b461... ✓ Open Forge openai deepseek-r1 +11 0e49122e76bd8b9ccb2fe10c0088c... Surplus Intel. openai 136+ models, 75 free 71e2ff34f8ef75... Prime Seed openai, anthropic claude-sonnet-4-6

What Each Column Means

ColumnWhat it tells you
Peer IDUnique identifier (use this for pinning via CLI)
NameProvider's display name
ProvidersAPI types they support (openai, anthropic, etc.)
ServicesModels they offer (+X means more not shown)
SessionsSuccessful transactions
GhostsSessions where they didn't deliver (lower = better)
VolumeTotal USDC they've processed

Viewing Peer Details

In AntStation Desktop: click on a peer to see all available models and pricing.

Via CLI (example below shows Dark Signal — same command works for any peer ID):

antseed network peer 4668854ba3e8b094e6f48fbeb59cec1cfde162f2

Output:

Providers & services openai-responses gpt-5.4 in $0.25/1M out $1.50/1M gpt-5.5 in $0.40/1M out $2.00/1M openai minimax-m2.5 in $0.21/1M out $0.84/1M minimax-m2.7 in $0.21/1M out $0.84/1M minimax-m2.7-highspeed in $0.42/1M out $1.68/1M

Part 4: Pinning a Peer (CLI Required)

The AntStation desktop UI is great for browsing, but pinning a peer requires the CLI. The desktop UI shows peers but doesn't let you pin one directly.

Install the AntSeed CLI

npm install -g @antseed/cli

Pin Surplus Intelligence (Recommended — Best Price/Quality)

Surplus Intelligence is my preferred peer: extremely competitive pricing for the model quality, plus 75+ free models for experimentation.

antseed buyer connection set --peer 0e49122e76bd8b9ccb2fe10c0088c41ceb608927

Pin a Different Peer

# Open Forge (best value on paid models)
antseed buyer connection set --peer 1d90f467689d499dc435e5744b4613c3203eb0aa

# Dark Signal (gpt-5.x, minimax)
antseed buyer connection set --peer 4668854ba3e8b094e6f48fbeb59cec1cfde162f2

# Prime Seed (Claude access)
antseed buyer connection set --peer 71e2ff34f8ef75...

Check Your Pin

antseed buyer status

Pinned peer                  │ 0e49122e76bd8b9ccb2fe10c0088c41ceb608927

Available Peers (As of This Guide)

PeerNameBest ForNotable Models
0e49122e76bd8b9... Surplus Intelligence Best price/quality (recommended) 136+ models, 75 free
4668854ba3e8b094... Dark Signal Chat, creative minimax-m2.7-highspeed, gpt-5.4
1d90f467689d499... Open Forge Value pricing deepseek-r1, deepseek-v3.2, qwen3-235b (free)
71e2ff34f8ef75... Prime Seed Claude access claude-sonnet-4-6

Part 5: Setting Up Hermes on Your VPS

Tell Hermes to Use AntStation

Give Hermes this prompt:

💬 Prompt for Hermes

I want to add AntStation (a peer-to-peer AI network buyer proxy) as a model provider option when I use the /model command.

The setup details:
- AntStation buyer proxy runs on localhost:8377
- It supports OpenAI Chat Completions API at http://localhost:8377/v1
- OPENAI_BASE_URL=http://localhost:8377/v1
- OPENAI_API_KEY=unused (any placeholder works)

Please:
1. Add "AntStation" as a selectable option in the /model command list
2. When selected, route all API calls through http://localhost:8377
3. Use the model name from the user's request verbatim (e.g., "minimax-m2.7-highspeed", "deepseek-v3.2")
4. Keep existing providers working alongside this new option
5. Save the configuration so it persists across sessions

Or manually add to ~/.hermes/config.yaml:

custom_providers:
  - name: antstation
    base_url: http://localhost:8377/v1
    api_key: unused
    api_mode: chat_completions

Update Your SSH Command

Add the AntStation port (8377) to your existing SSH command:

ssh -L 18789:localhost:18789 -L 9119:localhost:9119 -R 11434:localhost:11434 -R 8377:localhost:8377 account@your-vps-ip
PortDirectionPurpose
-L 18789Local → VPSHermes/OpenClaw dashboard
-L 9119Local → VPSOpenClaw web UI
-R 11434VPS → LocalAgent → Ollama
-R 8377VPS → LocalAgent → AntStation (NEW)

Verify the Tunnel is Active

From your Mac, confirm AntStation is running and the port is listening:

lsof -i :8377

You should see something like:

AntSeed   30575  ian  28u  IPv4  0xfb788224ec1ceaf9  0t0  TCP localhost:cruise-swroute (LISTEN)


Part 6: Switching Models in Hermes

How I Set It Up (No Pre-populated Model List)

I keep AntStation as a generic provider — no models listed in the config. Here's why:

Switching Models

Model and provider must be specified together — you can't switch the provider first and then pick a model in separate commands.

If you pinned Surplus Intelligence (recommended):

/model opus-4.7 --provider antstation
/model llama-3.3-70b-instruct --provider antstation
/model qwen2.5-72b --provider antstation
/model gpt-5.5 --provider antstation

If you pinned Dark Signal:

/model minimax-m2.7-highspeed --provider antstation
/model gpt-5.4 --provider antstation
/model gpt-5.5 --provider antstation

If you pinned Open Forge:

/model deepseek-r1 --provider antstation
/model deepseek-v3.2 --provider antstation
/model qwen3-235b-instruct --provider antstation

Switching to a New Peer

Want to try a different provider? Open Terminal on your Mac:

# See all peers
antseed network browse

# Pin your choice
antseed buyer connection set --peer <peer-id>

# Then use any model they offer in Hermes
/model <any-model-from-that-peer>

Part 7: How the Full Setup Works

┌─────────────────────────────────────────────────────────────────┐ │ My MacBook Pro │ │ ┌───────────────────────┐ ┌─────────────────────────────┐ │ │ │ AntStation Desktop │ │ Buyer Proxy (port 8377) │ │ │ │ App │ │ │ │ │ │ - Browse peers │ │ Pinned: Surplus Intel. │ │ │ │ - Chat with model │ │ │ │ │ │ - Manage payments │ │ │ │ │ └───────────────────────┘ └──────────────┬──────────────┘ │ └───────────────────────────────────────────────┼───────────────────┘ │ SSH Tunnel │ -R 8377:localhost:8377┌───────────────────────────────────────────────┼───────────────────┐ │ VPS (Ubuntu) │ │ │ ┌───────────────────────┐ │ │ │ │ Hermes Agent │───────────────────┘ │ │ │ Provider: antstation │ │ │ │ Model: minimax-m2.7 │ │ │ └───────────────────────┘ │ │ │ │ Routes through localhost:8377 ──────────────────────► │ └────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────────┐ │ Surplus Intelligence Provider (or whichever peer you've pinned) │ │ Returns AI response │ │ Payment settles in USDC on Base blockchain │ └─────────────────────────────────────────────────────────────────┘

Part 8: My Favorite Models by Provider

Surplus Intelligence (current pick — best price/quality)

My pinned peer. Extremely competitive on paid models and 75+ models free for experimentation. Run antseed network peer 0e49122e76bd8b9ccb2fe10c0088c41ceb608927 for the live price list — it changes faster than any blog post.

ModelTierBest For
opus-4.7Paid (premium reasoning)Complex reasoning, code, long-form writing
gpt-5.5Paid (often undercuts Dark Signal)Reasoning, multimodal
llama-3.3-70b-instructFREEGeneral chat
qwen2.5-72bFREECoding, long context

Dark Signal

ModelInput $/1MOutput $/1MBest For
minimax-m2.7-highspeed$0.42$1.68Fast creative writing
gpt-5.4$0.25$1.50General chat, code
gpt-5.5$0.40$2.00Reasoning, multimodal

Open Forge (best value)

ModelInput $/1MOutput $/1MBest For
qwen3-235b-instructFREEFree tier, long context
deepseek-v3.2$0.06$0.38General chat
deepseek-r1variesvariesDeep reasoning

Surplus Intelligence (cheapest)

ModelInput $/1MOutput $/1MBest For
75+ free modelsFREEExperimentation
Most paid modelsVery lowVery lowBudget use

Part 9: Troubleshooting

⚠️ Hermes Can't Validate the Model

You'll see:

Warning: could not reach the custom:antstation API to validate minimax-m2.7-highspeed.

This is cosmetic. Hermes tries to validate from the VPS before the tunnel is tested. Just send a message — if the tunnel is open, it works fine.

⚠️ "No Peer Pinned" Error

{"error":{"type":"no_peer_pinned",...}}

Fix — run on your Mac:

antseed buyer connection set --peer 0e49122e76bd8b9ccb2fe10c0088c41ceb608927

⚠️ Tunnel Not Working

From your VPS, test:

curl http://localhost:8377/v1/models

If it times out, reconnect your SSH session with the tunnel flag:

ssh -R 8377:localhost:8377 account@your-vps-ip

Part 10: Quick Reference

My Daily Workflow

  1. Open AntStation on Mac ✅
  2. Buyer proxy is already running in the background ✅
  3. SSH into VPS with tunnel:
    ssh -R 8377:localhost:8377 account@your-vps-ip
  4. In Hermes:
    /model gpt-5.5 --provider antstation
  5. Chat! 💬

Key Commands

<tr><td>Check balance</td><td><code>antseed buyer balance</code></td></tr>
TaskCommand
Browse peersantseed network browse
Pin a peerantseed buyer connection set --peer <id>
Check statusantseed buyer status
Manage fundsantseed payments
Switch peerantseed buyer connection set --peer <new-id>

Why I Love This Setup

See also: Want to understand the full AntSeed protocol — architecture, payments, reputation system, and why it matters? Read AntSeed: The Open Market for AI Inference That Has No Gatekeepers. Prefer a shorter, CLI-focused version? The P2P Quickstart covers just the essentials.

Happy chatting!

💬 Comments