# ARLI — full documentation for machines

> One OpenAI-compatible API for agents: 400+ frontier models from 59 providers
> via OpenRouter, a skill marketplace, and a hardened container sandbox — one API key.

Base URL: https://arliwork.com/v1
Docs: https://arliwork.com/docs
Site: https://arliwork.com
Marketplace: https://market.arliwork.com
Status: https://status.arliwork.com

---

## 1. What ARLI is

ARLI is an agent harness with a marketplace. It exposes frontier models through a
single OpenAI-compatible endpoint (routing handled by the platform, not by you),
and it lets agents install and buy skills from a marketplace. Agents can also run
inside a hardened container sandbox with metered compute.

Everything below is reachable with one API key and one base URL.

---

## 2. Register

    POST https://arliwork.com/api/register
    Content-Type: application/json

    {"email":"dev@company.com","password":"your_password","name":"Your Name"}

Response:

    {"ok":true,
     "api_key":"arli-…",        // platform key: full access to every /v1 route
     "litellm_key":"sk-…",      // LiteLLM virtual key: OpenAI-compatible model access
     "user":{"id":…,"email":"…","name":"…"},
     "credit_usd":1.0,
     "base_url":"https://arliwork.com/v1"}

Both keys work as Bearer tokens. New accounts start with **$1 of credit**, recorded
in the balance ledger (`GET /api/balance`). Keys are also visible at
https://arliwork.com/app/register.

Errors: `409` email already registered · `400` invalid email, password shorter than
6 characters, or a missing JSON body · `429` too many registrations from one IP.

---

## 3. Authentication

Send `Authorization: Bearer <key>` on every request. Two key types are accepted:

| Prefix | Type | Scope |
|---|---|---|
| `arli-` | Platform API key | full: models, agents, sandbox, account endpoints |
| `sk-` | LiteLLM key | model access only, OpenAI-compatible |

Requests without a Bearer token receive `401 {"detail":"Bearer token required"}`.
Requests with an unknown key receive `401 {"detail":"invalid token"}`.

---

## 4. Endpoints

### GET /v1/models

Lists the model ids you can call. OpenAI-shaped response:

    {"object":"list",
     "data":[{"id":"deepseek-v4-flash","object":"model","created":1789000000,"owned_by":"arli"}, ...],
     "passthrough":["openrouter/*"]}

`passthrough` lists routing patterns rather than selectable ids. Send a concrete
id from `data` as the `model` field of a completion. Requires a Bearer token.

Call the concrete id; the wildcard is a routing rule, not a model.

### POST /v1/chat/completions

OpenAI-compatible chat completion.

    curl https://arliwork.com/v1/chat/completions \
      -H "Authorization: Bearer YOUR_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{"model":"deepseek-v4-flash",
           "messages":[{"role":"user","content":"Hello"}]}'

Reasoning-capable models may return a `reasoning_content` field on the message
before or alongside `content`. Parse both; do not assume `content` is non-empty
when `finish_reason` is `length`.

Called without a `model` field, the endpoint falls back to the platform default.

### POST /v1/agent/run

Beta. Autonomous agent task execution: the platform plans, calls tools and
returns the result. Expect rough edges on first-generation agents; cap spend per
agent so a runaway loop cannot surprise you.

### GET /v1/agent/skills

Public, no key required. Returns the skill registry:

    [{"id":"code-review","name":"Code Review","tools":["read_file","search_files"]},
     {"id":"data-analysis","name":"Data Analysis","tools":["terminal","read_file","write_file"]},
     {"id":"research","name":"Deep Research","tools":["web_search","read_file","write_file"]},
     {"id":"devops","name":"DevOps","tools":["terminal","read_file"]}]

### POST /api/register

Developer registration. Returns the API key pair. See section 2.

---

## 5. OpenAI SDK compatibility

Drop-in replacement — change the base URL and keep everything else:

    # Python
    from openai import OpenAI
    client = OpenAI(api_key="YOUR_API_KEY", base_url="https://arliwork.com/v1")
    response = client.chat.completions.create(
        model="deepseek-v4-flash",
        messages=[{"role": "user", "content": "Hello"}],
    )

    # Node
    import OpenAI from "openai";
    const client = new OpenAI({ apiKey: "YOUR_API_KEY", baseURL: "https://arliwork.com/v1" });

---

## 6. Pricing

### Plans

| Plan | Price |
|---|---|
| Free | $0/mo |
| Pro | $29/mo |
| Builder | $99/mo |
| Team | $249/mo |
| Business | $10,000/yr — 25 seats, SSO, $5,000 usage |
| Enterprise | $50,000/yr — 100 seats, SAML and SCIM, DPA, $25,000 usage |
| Sovereign | $100,000/yr — two regions, residency by agreement, plus a $50,000 one-time deployment |

Prepaid commitments on the published rate: $10,000 at 12 percent off, $50,000 at 18 percent, $100,000 at 25 percent.

Industrial catalogue work is priced separately under Connect.ARLI. The whole list, platform and company plans
together: https://arliwork.com/pricing

### Model rates, per 1M tokens (input / output)

| Provider | Models | Input | Output |
|---|---|---|---|
| DeepSeek | v4-pro, v4-flash, r1, v3.2, chat | $0.10 – $0.70 | $0.20 – $2.50 |
| Qwen (Alibaba) | 3.7-max, 3.7-plus, 3-max, coder, coder-plus | $0.07 – $1.48 | $0.26 – $4.42 |
| GLM (Zhipu) | 5.2, 5.1, 5, 4.7, 4.6, vision | $0.06 – $0.97 | $0.40 – $3.04 |
| Kimi (Moonshot) | k3, k2.7-code, k2.6, k2.5 | $0.57 – $3.00 | $2.30 – $15.00 |
| MiniMax | m3, m2.7, m2.5, m1 | $0.15 – $0.55 | $0.90 – $2.20 |
| ByteDance (Seed) | 2.0-lite, 1.6, 1.6-flash | $0.07 – $0.25 | $0.30 – $2.00 |
| Baidu (ERNIE) | 4.5 | $0.42 | $1.25 |
| Tencent (Hunyuan) | hy3, a13b | $0.06 – $0.20 | $0.21 – $0.80 |
| Xiaomi (Mimo) | v2.5-pro, v2.5 | $0.14 – $0.44 | $0.28 – $0.87 |
| Kwaipilot | kat-coder-pro-v2.5, v2 | $0.30 – $0.74 | $1.20 – $2.96 |
| Inception | mercury-2 | $0.25 | $0.75 |
| Aion Labs | 3.0 | $3.00 | $6.00 |
| StepFun | 3.7-flash | $0.20 | $1.15 |

### Sandbox compute

- vCPU: $0.000004 per second per core ($0.0144/hr)
- RAM: $0.000002 per second per GiB ($0.0072/hr)
- Storage: 60 GB free, then $0.00007 per GB per hour
- Agent run: 10 credits · Chat: 1 credit per 1K tokens

Example costs: 1 core + 512 MB for 5 minutes = $0.0014 · 2 cores + 1 GB for one
hour = $0.036 · 8 cores + 8 GB multi-agent = $0.173/hr.

---

## 7. First day, the supported path

- 0–5 min: register, collect the key, store it in a vault. $1 free credit.
- 5–20 min: run the curl quickstart above. If it answers, endpoint and auth are correct.
- Day 1: try one marketplace skill on a low-risk task before wiring it into production.
- Day 1–3: set a token or credit cap per agent.
- Week 1: report bugs, edge cases and billing questions — fixes ship weekly.

---

## 8. Support

- Developer: hq@arliwork.com
- Enterprise: enterprise@arliwork.com
- Telegram: https://t.me/arliwork
- Status: https://status.arliwork.com
- Changelog and docs: https://arliwork.com/docs

## 9. Other machine-readable files

- https://arliwork.com/llms.txt — short index (start here)
- https://arliwork.com/discovery.json — the same facts as JSON
- https://arliwork.com/sitemap.xml — public pages