> For the complete documentation index, see [llms.txt](https://docs.palliora.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.palliora.org/ai/build/chat-completions-service.md).

# Build a chat-completions-compatible service

Most AI tooling (chat UIs, agent frameworks, coding assistants) already speaks one API shape: `POST /v1/chat/completions` with a standard chat-completions body. You don't need any of those tools to know Palliora exists. Wrap a Compute Contract behind that same shape, and anything that speaks that API can call Palliora instead, unmodified.

This is the same pattern a few real Palliora sample services use internally. It's worth building your own small version to understand the shape, rather than only reading about it.

## The core idea

An HTTP request comes in shaped like a standard chat completion. Your service:

1. Builds the same JSON payload [Quickstart](https://github.com/palliora-org/palliora-docs-with-ai/tree/main/build/quickstart.md) sends through `inferenceCompute`, using the request body's `model` and `messages` fields directly.
2. Submits the Compute Contract.
3. Polls for the result, the same `scanForBlockEvent` / `fetchAndDecodeExtrinsic` pattern from [Call a model](/ai/build/call-a-model.md).
4. Wraps the answer back into a chat-completions-shaped response object and returns it.

The caller never sees a block number, an agreement ID, or a Guardian. They see the same response shape they'd get from any other chat-completions endpoint.

## A minimal version

```ts
// server.mjs
import express from "express";
import {
  init,
  getKeyring,
  getGuardianAddress,
  inferenceCompute,
  getApi,
  scanForBlockEvent,
  fetchAndDecodeExtrinsic,
} from "@palliora.org/chainsdk";

init({ pallioraWs: process.env.PALLIORA_WS ?? "wss://staging.manas-rpc.palliora.org" });

const keyring = await getKeyring();
const account = keyring.addFromUri("//Bob"); // dev-only, replace for real use

const app = express();
app.use(express.json());

app.post("/v1/chat/completions", async (req, res) => {
  try {
    const { model, messages } = req.body;

    const guardians = (await getGuardianAddress()).slice(0, 3).map((g) => g.address);
    const result = await inferenceCompute(
      {
        input: JSON.stringify({ model, messages, stream: false }),
        guardians,
        fee: { amount: 0.05, computeRate: 0.00002 },
        deadline: 0,
      },
      account,
    );

    const api = await getApi();
    const match = await scanForBlockEvent(
      api,
      {
        predicate: async (block, _event, phase) => {
          if (!block || !phase.isApplyExtrinsic) return false;
          const ext = block.block.extrinsics[phase.asApplyExtrinsic.toNumber()];
          if (ext?.method?.section?.toLowerCase() !== "compute" || ext?.method?.method?.toLowerCase() !== "result") {
            return false;
          }
          const emitted = "0x" + Array.from(ext.method.args[0]).map((b) => ("0" + (b & 0xff).toString(16)).slice(-2)).join("");
          return emitted === result.agreementId;
        },
      },
      result.blockNumber + 1,
      300,
    );

    const { decoded } = await fetchAndDecodeExtrinsic(match.blockNumber, match.extrinsicIndex ?? 0);
    const answer = new TextDecoder().decode(
      Buffer.from(decoded.method.args.contract.compute.input.Inline.data.slice(2), "hex"),
    );

    res.json({
      id: result.agreementId,
      object: "chat.completion",
      model,
      choices: [{ index: 0, message: { role: "assistant", content: answer }, finish_reason: "stop" }],
    });
  } catch (err) {
    res.status(500).json({ error: err.message });
  }
});

app.listen(3000, () => console.log("Listening on :3000"));
```

Point any chat-completions-compatible client at `http://localhost:3000/v1` and it works without knowing what's coordinating it underneath.

## Where this needs to grow up before production

This minimal version is plaintext, uses a single dev signer for every request, blocks the HTTP response on a network round trip that can take a while, and has no retry or timeout handling beyond the 300-block window. A real service built this way typically needs:

* **Per-user signing.** One shared dev key means every request is attributed to the same account. A real service signs each user's compute agreement with that user's own key.
* **Privacy, if it matters for your use case.** This example never encrypts the prompt or the answer, see [Encrypted Compute](https://github.com/palliora-org/palliora-docs-with-ai/tree/main/build/guides/encrypted-compute.md) for the threshold-encrypted path.
* **Async handling.** A network round trip is not instant. Streaming responses, request queues, or a webhook-style callback all fit this better than a single blocking HTTP call for anything beyond a demonstration.
* **Tool calls.** If the thing behind your endpoint needs to call external tools (web search, a calculator, another API), each tool call is its own Compute Contract in the same pattern, triggered by what the model returns, same as a normal chat-completions tool-calling loop, just with each step running as a Palliora compute agreement instead of a direct API call.

See [Palliora AI](https://github.com/palliora-org/palliora-docs-with-ai/tree/main/products/palliora-ai.md) for what a fuller version of this pattern looks like once all of that is addressed.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.palliora.org/ai/build/chat-completions-service.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.