v1.0.11.0 is here: six tools, one workspace.See what's new →
Armor EdgeAI
EN
Download

API

Your GPU,
for any tool

The AI Chat module ships an OpenAI-compatible API — serve your local compute to third-party applications: terminal coding agents, editor plugins, automation scripts. Zero API fees, no queues, and your code and prompts never leave the machine.

Your computer
Terminal coding agentsEditor pluginsAutomation scripts
no key needed
Armor EdgeAI127.0.0.1:2918 · /v1 · /anthropic/v1
Devices on your local networkOff until you turn it on; an API key is required
InternetNo direct access; never expose it to the public internet

01 — How It Works

While the app runs, the API is there

Once a model is running, the chat module doubles as a local inference service — any tool that supports a custom endpoint can connect directly.

/v1

OpenAI-compatible

Standard chat-completions endpoint (/v1) with SSE streaming — tools written for the OpenAI API mostly just work.

/anthropic/v1

Anthropic-compatible

An Anthropic-compatible endpoint (/anthropic/v1) is also provided, so tools built for the Claude API connect too.

:2918

Local endpoint

The service runs on port 2918 of your computer; by default it binds to localhost only (127.0.0.1) — data stays on your machine.

02 — Setup

Connect in three steps

  1. 1

    Start a model in the app

    Open AI Chat, pick and launch a downloaded model — it becomes the inference engine behind the API.

  2. 2

    Get the endpoint details

    The "API Configuration" page in the sidebar lists the Base URL and available model names — copy them.

  3. 3

    Paste into your tool

    Enter the Base URL and model name in your tool's custom-provider settings — done. Inference now runs on your own GPU.

The API page in the app: connection details, API keys and usage
The API page in the app: the URL and model ID to enter in your tool, API keys, and usage per caller.

03 — Example

Example: coding in the terminal with opencode

opencode is an open-source terminal AI coding agent. Point it at the local endpoint and your coding assistant runs entirely on your own GPU — reading code, editing code, running commands, fully offline.

In opencode's custom-provider settings, you only need three pieces of information:

Base URL
http://127.0.0.1:2918/v1
API Key
same computer: any string (most tools require a value); devices on your local network: a key created on the "API Configuration" page
Model
a model name as listed on the app's "API Configuration" page

For the exact configuration format, follow the custom-provider section of the opencode documentation; the endpoint and model names shown in the app's "API Configuration" page are authoritative. Any tool that supports a custom OpenAI-compatible provider — editor plugins, chat clients, scripts — needs the same three pieces.

curl http://127.0.0.1:2918/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer local" \
  -d '{
    "model": "<model ID from the API page>",
    "stream": true,
    "messages": [
      { "role": "user", "content": "Explain what this function does in three points" }
    ]
  }'

Programs on the same computer can use any string as the key; devices on your local network must use an armor_sk_ key you created.

03+ — More Ways To Connect

Not just coding tools

Phones and tablets connect with a quick scan, and AI agents can use this computer's text recognition and speech transcription directly.

04 — Capabilities

Your tools get a complete inference service

Streaming responsesSSE streaming works as usual — your coding agent's token-by-token output feels identical to a cloud API.
Context & parametersContext length and generation parameters depend on the selected model — the same engine as in-app chat.
No quota anxietyNo token billing, no rate-limit tiers — the ceiling is your hardware.

Before you start

LAN mode

By default the API binds to localhost only. To let other computers on the same network use it, turn on local network access in Settings; other devices must send an API key you create on the "API Configuration" page to make calls. The connection itself is unencrypted HTTP, so enable it only on trusted networks and never expose it to the public internet.

AI agent connections (MCP)

Besides the chat endpoints, there is a separate connection for AI agents: it lets tools you trust call text recognition and speech transcription. It is off by default; turn it on under "Settings → AI Agent Connections" and create a connection for each tool.

Performance is your hardware

Response speed and usable context length are determined by your GPU and the selected model.

Terms still apply

Use via the API remains subject to the EULA and the Acceptable Use Policy.

05 — Use Cases

Use Cases

Full write-ups from setup to finished result — real machine, real tool, with the shortcomings written down alongside the wins.

Read the case →
The finished case: a web piano built on a local modelA web piano, built by OpenCode on your own GPU

Next

Get to know the Chat module, then connect your tools