Connect Your Agent to Morpheus: Pick Your Agent, Then Your Route

At the Morpheus town hall on 24 September 2026, one of the contributors asked for a page where “I can just pick my agent” and go straight to “the simple setup” for connecting it. This is that page.

Start with the route, then find your agent. Commands and config files are tucked into panels you can open when you need them.

How It Works

An AI agent is a program that does tasks for you, and every few seconds it asks an AI model what to do next. Normally those questions go to a company like OpenAI or Anthropic. Morpheus swaps that company for an open marketplace of independent providers who run the models and get paid in MOR.

Your agent can’t tell the difference, because Morpheus speaks the same API language as OpenAI. Most agent apps support this, usually under a setting called “custom endpoint” or “OpenAI-compatible provider”.

Choose Your Route

There are four ways in. The ones on the left are quicker to start; the ones on the right keep more of the chain in your own hands.

Not sure? Here’s how we’d choose:

• Just curious? Use the hosted gateway. You’ll be running in minutes, with no crypto needed.
• Want an agent without the setup? A managed agent is the easiest start, but keep private material out of it until its prompt path is documented.
• Using an agent every day? Run your own Uplink. It’s the route we use.
• Want nothing between you and the provider? Run your own node.

Who reads your prompts? On the first two routes, a company sits in the middle. Morpheus’s own privacy explainer says the gateway runs under a no-logging policy, yet admits that “the gateway system could log your prompts” and that it can’t prove otherwise yet. On every route, the provider running the model sees your prompt too, unless it runs inside a sealed, verifiable chip called a TEETEETrusted Execution Environment. A hardware-secured region of a CPU or GPU where code runs in isolation, so even the machine's operator can't read what's happening inside. TEEs give decentralised AI inference privacy guarantees.Like a bank vault inside a bank. The bank owns the building, staffs the lobby, and runs the security cameras. But what's inside the vault is invisible to everyone, including the bank staff, unless the customer opens it.TEE in the glossary →.

The managed route is the hardest to see into. We read the open-source code behind installopenclaw.xyz: questions leave your agent through the operator’s own relay, a “Central Inference Gateway” whose code isn’t published. The operator’s notes point it at Morpheus, but nothing says where it forwards prompts or whether it keeps them. We asked the builder in a public GitHub issue on 29 September 2026 and will update this page with the answer.

Find Your Agent

Each card shows which routes the agent supports and whether we tested it ourselves. Click through for its setup.

Is your app missing? If its settings let you change the “base URL” or add a custom OpenAI provider, it will work with the hosted gateway or your own Uplink.

Set It Up

For your own Uplink, every agent needs the same three things, all shown in Uplink’s console:

1. The web address of your Uplink, ending in /v1.
2. A key for that agent. Make a separate one for each agent, so you can switch one off without touching the rest.
3. The model’s ID code, a long string starting with 0x. Use the code, never the model’s name. Some names are shared by several listings, and Uplink can pick one that locks up far more MOR (issue 41).

openclaw --profile uplink setup --non-interactive --accept-risk \
  --auth-choice custom-api-key \
  --custom-base-url https://YOUR_UPLINK_HOST/v1 \
  --custom-api-key sk-YOUR_AGENT_KEY \
  --custom-model-id 0xYOUR_MODEL_ID \
  --custom-provider-id uplink \
  --custom-compatibility openai --custom-text-input

export UPLINK_KEY=sk-YOUR_AGENT_KEY
hermes config set model.provider custom
hermes config set model.base_url https://YOUR_UPLINK_HOST/v1
hermes config set model.default 0xYOUR_MODEL_ID
hermes config set model.key_env UPLINK_KEY

What Happened When We Tried It

We gave each agent a small job that forced it to use a tool: work out a SHA-256 fingerprint of a piece of text using its code tool. All three came back with the same correct answer.

For comparison, we ran Agent Zero’s two tasks on Venice the next day: 77 and 54 seconds, both correct, with another session sharing the machine. One run each proves nothing about speed, except that Morpheus didn’t slow the agent down. Hermes hit one snag of its own: its safety rules blocked a quick Python command, so it wrote a script file and ran that instead.

Fact: A one-line question turned into two to four requests to the model, each carrying the agent’s full instruction manual.

Take: Agents are hungry customers. On the hosted gateway you pay per token, so that overhead lands on your bill; judge models by the cost of a finished task. On your own Uplink the deposit depends on how long the session runs and the provider’s price, and the size of the prompt doesn’t change it.

Your MOR Is a Deposit, If One Setting Is Right

On your own Uplink or node, MOR works as a deposit. You lock some while a session runs, and the protocol pays the provider from its own pool.

That’s only true while a session setting called direct payment stays off. Switch it on, and the Morpheus contract keeps part of your deposit as payment. In Uplink the setting is SESSION_DIRECT_PAYMENT, and its code defaults it to off. Leave it alone: all eight recent sessions from our Uplink and node showed it off on the blockchain when we checked on 29 September 2026.

Four Tips Before You Start

• Every key can lock up your MOR. Anyone holding an agent’s key can open sessions against your wallet, so keep keys out of chats and screenshots.
• Pay for a proper blockchain connection. On a free public connection, our node missed confirmations for sessions that had opened, retried, and reported errors for things that had worked. Check your sessions on the blockchain before you retry a “failed” one.
• Uplink’s usage page shows zero tokens for most agents. Version 1.0.16 doesn’t count streamed replies, and every agent above streams. Our figures come from a patched copy; we’ve offered the fix to the Uplink team (issue 42).
• Test with a strong model. Small, cheap models are fine for checking the plumbing, but judge Morpheus on a capable one.

What’s Still Missing

On 29 September 2026 the Grok plug-in was an early version outside xAI’s plug-in store, so Grok users install it by hand. Morpheus’s Eliza guide is a placeholder, and nothing in Hermes’s docs mentions Morpheus.

The bigger gap is privacy on managed agents. We think a free hosted agent is how most newcomers will meet Morpheus, and today what happens to their prompts rests on a relay nobody can inspect. For anything you wouldn’t paste into a public chat, I’d use your own Uplink or node. For everything else, the hosted gateway is a perfectly good way in.