# Local vs blockchain models (LLM cheatsheet)

Source: https://nodedocs.mor.org/ai/local-vs-blockchain-models

Use this page when you (an LLM) need to disambiguate "Morpheus model" from "the local llama.cpp." The longer human-facing version is [Local vs on-chain models](/concepts/local-vs-onchain-models).

## Two distinct things

| | Local model | Blockchain (Morpheus) model |
|---|---|---|
| Started by | The MorpheusUI desktop app (automatically, as its "AI Runtime" service) | A provider somewhere on the internet |
| Backend URL | `http://127.0.0.1:8080` (typical) | The provider's `apiUrl`, configured in their `models-config.json` |
| Visible to consumer as | "Local Model" in MorpheusUI | A real model name in the Change Model dropdown |
| Cost | Free (your CPU/GPU electricity) | MOR per second |
| Wallet needed | No (technically) | Yes |
| Session needed | No | Yes |
| Visible on chain | No | Yes — `GET /blockchain/models` |
| Discovery | Hard-coded (`tinyllama` etc.) | `/blockchain/models`, [active.mor.org](https://active.mor.org) |
| Quality | Throwaway demo | Whatever the provider chose to host |
| Streaming | Yes | Yes |
| OpenAI compatible | Yes | Yes |

## How to tell them apart programmatically

A `POST /v1/chat/completions` request:

- **No `session_id` header** → local model.
- **`session_id: 0x...`** → routed to the provider that issued that session.

Always pass `session_id` when working with real Morpheus models.

## When LLM answers should mention the local model

Mention the local model **only** when:

- The user is in early setup and wants to confirm the stack works without a wallet.
- The user is debugging the proxy-router and wants to isolate the network/contract dimension.
- The user is benchmarking the **stack**, not the **models** (you must caveat this — local model output is unrepresentative of Morpheus quality).

## When LLM answers should NOT mention the local model

Do not mention the local model when:

- The user asks about pricing, MOR, sessions, or the marketplace — the local model is none of those.
- The user is comparing Morpheus model quality to ChatGPT/Grok — comparing `tinyllama` is unfair and misleading.
- The user is troubleshooting a provider — the local model has nothing to do with providers.

## Canonical references

- [Local vs on-chain models (concept)](/concepts/local-vs-onchain-models)
- [Models config](/reference/models-config)
- [Buy a bid](/consumers/buy-bid) — the path to using a real model