Agent Pods — Overview - x402 Singularity Layer

Agent Pods are always-on hosted AI agents. Deploy one in a couple of minutes and it stays online 24/7 — you chat with it from the dashboard and on Telegram or Discord. Today you deploy a ClawPod (the OpenClaw agent) on its own dedicated CPU machine. A HermPod (the Hermes self-improving agent) is coming soon. Each pod comes with its own crypto wallet, memory, the x402-compute + x402-layer skills, and the native Singularity MCP preinstalled — so the agent is economically active out of the box: it can buy confidential compute, use the marketplace, and pay any x402 endpoint itself, all under caps you control.

In this section

Deploy a Pod

Agent, Managed vs BYOK, machine, channels, pay, free trial.

Manage a Pod

Workspace tabs, lifecycle, auto-renew & grace.

Models

Tiers, uncensored models, /model, the Model setting.

Pod Wallet

Coinbase-secured wallet, spend controls, x402 pay.

Memory

Raw, Lossless Context, semantic memory.

Scheduler

Scheduled tasks, delivery to a channel.

OpenAI access

Call the agent from any OpenAI client.

API reference

For developers & agents — every endpoint, auth, and schema.

What you get

An always-on agent — hosted on a dedicated machine, reachable from the dashboard and your chat apps.
Managed or BYOK AI — we run a model from your tier (billed from your credits), or bring your own OpenAI-compatible key + model.
Its own wallet — a Coinbase-secured wallet (Base + Solana) with keys in a secure enclave. Fund it and the agent can pay for things; withdraw anytime.
Preinstalled skills + MCP — x402-compute, x402-layer, and the Singularity MCP (marketplace / Slayer tools), auto-wired to your account under revocable, capped credentials.
Memory — persistent per-pod memory so the agent remembers across conversations.

Agents

Agent	Framework	Status
ClawPod	OpenClaw	Available now
HermPod	Hermes (self-improving, grows its own skills)	Coming soon

Deploy

Pick an agent + AI mode

Managed — we run the LLM. Choose a tier: Starter (chat), Pro or Max (add computer-use + memory). Each tier bundles a matched machine and a curated menu of models the agent can switch between with /model.BYOK — bring your own OpenAI-compatible key and model, and pick any machine. You pay for the machine + a small service fee; your AI runs on your key.

Add channels (optional)

Paste a Telegram or Discord bot token to reach your agent there. (Slack, WhatsApp and Signal are coming soon.) You can add or change channels anytime after deploy — or just use the in-dashboard Chat.

Pay

From your platform credits or per-month USDC (x402). Or start on the free 24h trial (Starter/Pro) — no upfront payment.

Tiers & models (managed)

Each managed tier bundles a matched machine floor with a curated list of models — all included. Your agent starts on the tier default and can switch to any model in its tier anytime with /model <name> in chat, no redeploy. The managed proxy meters real usage per call, so switching is cost-safe, and a pod can never reach an off-tier (pricier) model.

Starter
Pro
Max

Chat agent on Telegram or Discord. Fast, cheap, text-only (no computer use). Runs on a 2 GB machine.

deepseek/deepseek-chat-v3.1 (default)
qwen/qwen3-235b-a22b-2507
z-ai/glm-4.5-air
openai/gpt-5-mini

Adds computer use — the agent can see screenshots and drive a real browser. Runs on a 4 GB machine.

google/gemini-2.5-flash (default)
google/gemini-2.5-flash-lite
z-ai/glm-4.6
openai/gpt-4.1-mini

Top reasoning + reliable computer use for heavy multi-step agents. Runs on an 8 GB machine.

google/gemini-2.5-pro (default)
anthropic/claude-sonnet-4.5
openai/gpt-5.1
google/gemini-2.5-flash

BYOK ignores tiers entirely — you bring any OpenAI-compatible model and pick any machine size.

Free 24h trial

Starter and Pro can deploy on a free 24-hour trial — a grant-funded pod with no upfront payment. It runs exactly like a paid pod, then is auto-destroyed at expiry (trials never auto-renew). Keep it by converting to a paid plan before the countdown ends. The AI allowance on a trial is capped, and the total giveaway is hard-capped by the trial campaign budget.

Billing — three separate rails

A pod’s spending flows through three independent, individually-capped rails. Keeping them separate is what makes an always-on agent safe to leave running.

1. Managed inference (the LLM)

Every managed pod ships with a free included AI allowance bundled into the plan. Once that’s used up, inference overage bills from the owner’s Singularity credits, bounded by a ~$20/day per-pod AI cap. At zero balance or over the daily cap, managed inference fails closed (the agent pauses thinking rather than overspending). BYOK pods skip this rail — their AI runs on your own key.

2. Delegated grid-compute budget

So the agent can use Singularity Compute + the x402 marketplace out of the box, each pod gets a pod-scoped, daily-capped Compute key (default-deny route allow-list). The budget defaults to $2/day** and you can raise it up to a **$ 100/day ceiling from the pod’s settings — or swap in your own key for no ceiling. This key is scoped to your account and revocable anytime.

3. The pod's crypto wallet

The agent’s own wallet (Base + Solana) funds x402 purchases and on-chain sends. Agent spending is off by default — you arm it and set a per-transaction cap and a rolling-24h cap (bounded by a platform ceiling). Your own withdrawals are uncapped.

Wallet

Every pod gets its own wallet (EVM on Base + Solana). Private keys live in Coinbase’s secure enclave — never on the pod, never with us. Fund it (deposit) so the agent can buy any x402 endpoint, pay for APIs, and take on-chain actions. You can withdraw or send anytime from the Wallet tab. Agent spend is capped and off by default (arm it and set a per-transaction limit in Wallet → Spending controls); your own withdrawals are uncapped. Funds are never lost on destroy — the wallet stays reachable under Archived.

Skills & MCP

Every pod ships economically active:

x402-compute — buy confidential compute on the Singularity grid.
x402-layer — pay x402 endpoints and manage Layer operations.
Singularity MCP — native marketplace / Slayer tools via tool-calling.

These are auto-wired to your account using the pod’s delegated Compute key + a scoped Studio PAT — capped and revocable. The agent also ships 50+ built-in OpenClaw skills, installable from ClawHub.

Memory

Every pod keeps persistent per-pod memory so the agent remembers across conversations.

Starter — on-pod memory files (private to the pod).
Managed Pro / Max — enhanced cross-session recall; Max adds mem0 semantic memory, brokered so the key never touches the VM. (Advanced semantic recall is rolling out — some higher-tier memory features may activate progressively.)
BYOK — on-pod memory files by default, or bring your own mem0 key.

Channels & pairing

When you add a channel, message your bot once — it replies with a pairing code. Paste that code in the pod’s Channels tab to let yourself in. Your agent stays private until you pair.

Lifecycle

Auto-renew — a paid pod is prepaid for a runtime window (e.g. 1 month) and auto-renews from your credits at expiry, holding in a short grace window if you’re briefly short before destroying. Trials never renew.
Extend — top up the runtime window early.
Manage — restart, redeploy, stop, or destroy from the pod detail page. Destroying refunds remaining prepaid time and releases the machine (wallet funds are preserved).
Update agent — pull the latest agent features/skills onto a running pod.
Heartbeat, diagnostics & logs — the pod heartbeats its status; run Diagnose or pull Logs from the dashboard when you need to see what it’s doing.

Deploy programmatically (agentic)

Pods are fully API-driven, so an agent can deploy and manage other pods. Auth with an X-API-Key, a signed compute session, or x402; pay with credits (use_credits: true) or x402.

# Deploy a managed Pro pod from credits
curl -s -X POST https://compute.x402layer.cc/pods \
  -H "X-API-Key: $COMPUTE_API_KEY" -H "Content-Type: application/json" \
  -d '{"agent_id":"openclaw","ai_mode":"managed","tier":"pro",
       "plan":"<plan_id>","prepaid_hours":720,
       "channels":{"telegram":"<bot_token>"},"use_credits":true}'

# List + manage
curl -s https://compute.x402layer.cc/pods           -H "X-API-Key: $COMPUTE_API_KEY"
curl -s https://compute.x402layer.cc/pods/<id>      -H "X-API-Key: $COMPUTE_API_KEY"
curl -s -X POST https://compute.x402layer.cc/pods/<id>/actions \
  -H "X-API-Key: $COMPUTE_API_KEY" -H "Content-Type: application/json" -d '{"action":"restart"}'

See the public catalog for agents, tiers, models, and plans: GET https://compute.x402layer.cc/pods/catalog. The x402-compute skill documents the full flow for agents.

Overview

User Guide

Slayer Agent

Core flows

Staking Engine

Singularity Cloud Network

Contract Addresses

ERC8004

Skills

Singularity Payments

Agentic Fundraisers

Enterprise

Agent Pods — Overview

In this section

Deploy a Pod

Manage a Pod

Models

Pod Wallet

Memory

Scheduler

OpenAI access

API reference

What you get

Agents

Deploy

Tiers & models (managed)

Free 24h trial

Billing — three separate rails

1. Managed inference (the LLM)

2. Delegated grid-compute budget

3. The pod's crypto wallet

Wallet

Skills & MCP

Memory

Channels & pairing

Lifecycle

Deploy programmatically (agentic)

​In this section

Deploy a Pod

Manage a Pod

Models

Pod Wallet

Memory

Scheduler

OpenAI access

API reference

​What you get

​Agents

​Deploy

​Tiers & models (managed)

​Free 24h trial

​Billing — three separate rails

1. Managed inference (the LLM)

2. Delegated grid-compute budget

3. The pod's crypto wallet

​Wallet

​Skills & MCP

​Memory

​Channels & pairing

​Lifecycle

​Deploy programmatically (agentic)

In this section

What you get

Agents

Deploy

Tiers & models (managed)

Free 24h trial

Billing — three separate rails

Wallet

Skills & MCP

Memory

Channels & pairing

Lifecycle

Deploy programmatically (agentic)