Connect AutoClaw to claudeapi.com: Get Claude Opus 4.7 Running in Your Local Agent in 5 Minutes

AutoClaw is a desktop Agent application with a built-in local Gateway. It ships with multiple Agent templates — AutoClaw, DeepResearch, Cron, Browser Use — and supports MCP Servers, a Skills Store, plus connections to Telegram, WhatsApp, Lark, and other messaging platforms. Out of the box it offers Auto and GLM-5-Turbo as built-in models, but to get the best results — especially for complex planning, code generation, and long-context research — plugging in an external Claude Opus 4.7 is the most impactful upgrade you can make.

This guide follows the exact UI flow: Settings → Models & API → Add Custom Model → Fill in the form → Connectivity Test → Save — every field covered, nothing left to guess. By the end, you’ll be chatting with Claude.

Applies to: AutoClaw current release (with Settings/Models & API module and Hermes Evolution Board).

1. Before You Start: Three Things You’ll Need

Save these parameters somewhere handy so you’re not switching between windows:

Provider:        Custom
API Protocol:    Anthropic
Base URL:        https://gw.claudeapi.com
Model ID:        claude-opus-4-7
Display Name:    claude-opus-4-7
Context Window:  200000
Max Output:      32000

Provider:        Custom
API Protocol:    Anthropic
Base URL:        https://gw.claudeapi.com
Model ID:        claude-opus-4-7
Display Name:    claude-opus-4-7
Context Window:  200000
Max Output:      32000

2. Step 1: Open Settings

Click Settings (the gear + slider icon) in the bottom-left corner of AutoClaw’s main interface.

You’ll also see Feedback on the same row. Settings is the gear-shaped one.

3. Step 2: Navigate to Models & API

Select Models & API from the left sidebar.

This page has two sections:

• Built-in Models: Auto and GLM-5-Turbo — ready to use with no configuration.
• Custom Models: Where you add external models. Look for the Add Custom Model button on the right.

Claude goes into the Custom Models section. Click Add Custom Model.

At the bottom of the page you’ll see Gateway URL showing Connected, with a default port of ws://127.0.0.1:18789. AutoClaw communicates with external models through this local Gateway — it must show Connected before you proceed. If it shows Disconnected, click Reconnect in the upper right first.

4. Step 3: Fill In the Add Model Form (The Core Step)

The Add Model dialog contains the following fields. Fill them in exactly as shown:

The yellow notice at the top reads: “By adding an external model, you acknowledge and agree to use it at your own risk.” This is AutoClaw’s standard disclaimer for all external models — just acknowledge and proceed.

What About Other Claude Models?

To add a different Claude model, just swap Model ID and Display Name to the corresponding string — everything else stays the same:

💡 How to choose: claude-sonnet-4-6 offers the best price-to-performance for daily use, claude-opus-4-7 for complex planning and long code generation, claude-haiku-4-5-20251001 for high-frequency routing and classification tasks.

5. Step 4: Run the Connectivity Test (Strongly Recommended)

After filling in the form, don’t hit Save yet — click the Connectivity Test button at the bottom.

The button description reads:

Not tested yet — Sends a real request using the current config to verify the endpoint.

It fires an actual API request using your configuration. Here’s what to expect:

Once the test passes, click Save to finish.

6. Step 5: Set Claude Opus 4.7 as the Default Model

Back on the Models & API page, you’ll see claude-opus-4-7 in the Custom Models section with three actions:

• Set Default: Make it the default model for all conversations
• Edit: Modify configuration (key rotation, parameter adjustments, etc.)
• Delete: Remove the model

Click Set Default to make AutoClaw use Claude Opus 4.7 by default across all Agents.

7. Step 6: Go Back to the Chat Interface and Start Talking

Click ← Back to App in the upper left to return to the main interface.

Key things to verify:

1. Model selector at the bottom shows claude-opus-4-7 (click the dropdown to switch between added models)
2. Capability bar on the right reads Hermes Evolution · 100% — this is AutoClaw’s intelligence evolution indicator; full bar means you’re running the strongest configuration
3. Top-right corner shows your account quota (e.g., 2,300 | Purchase)

Send a quick test message:

hi

hi

You should immediately get a response from Claude (something like “Hi! How can I help you today?”). When you see the Task ended label, AutoClaw has detected that the interaction round is complete.

8. Going Further: Feed Claude to All Your Agents

AutoClaw’s Agents tab on the left ships with four templates, each configurable with its own default model:

You can also click + Create Agent at the top to create new Agents with individually assigned models — for example, Cron tasks on Sonnet 4.6 to save costs, while DeepResearch runs on Opus 4.7 for maximum quality.

Want to Save More? Use a Tiered Strategy

High-frequency routing / simple classification  →  claude-haiku-4-5-20251001
Everyday chat / content generation              →  claude-sonnet-4-6
Complex planning / critical decisions           →  claude-opus-4-7

High-frequency routing / simple classification  →  claude-haiku-4-5-20251001
Everyday chat / content generation              →  claude-sonnet-4-6
Complex planning / critical decisions           →  claude-opus-4-7

claudeapi.com pricing across three tiers (per million tokens):

With a “70% Haiku + 25% Sonnet + 5% Opus” split, you can cut costs by over 50% compared to running everything on Sonnet.

9. Connecting IM Channels & Skills

The main interface has two quick-access cards:

• Quick Setup: Set your name and role so AutoClaw better understands your context
• Connect IM: Hook your Agent into Telegram, WhatsApp, Lark, and other messaging platforms — chat with Claude right inside your favorite messenger

The Skills Store in the bottom left lets you mount skill packs to extend capabilities. These all share the Claude model you configured above — set it up once, use it everywhere.

10. Troubleshooting

Issue 1: Gateway URL Shows Disconnected

The three buttons at the bottom of the page, in order of escalation:

⚠️ Use Diagnose with caution: It may wipe your manually added Custom Models. Back up your configuration screenshots first.

Issue 2: Connectivity Test Keeps Returning 401

Check these in order:

1. Is the API Key fully copied? (No leading/trailing spaces or missing characters)
2. Is the key still in “Enabled” status? (Console → Key Management)
3. Does the account have sufficient balance?
4. Is the Base URL exactly https://gw.claudeapi.com? (Do not use gw.claudeapi.com/v1 — that’s the OpenAI-compatible path, but AutoClaw uses the native Anthropic protocol here)

Issue 3: Model Not Visible in the Selector After Saving

Click Reconnect in the upper right of the Models & API page, then go back to the main interface and open the model dropdown — it should appear.

Issue 4: Conversation Error “context length exceeded”

Edit the model and increase the Context Window value. Opus 4.7 / Sonnet 4.6 support up to 1M (1000000). Note that Opus 4.7 uses a newer tokenizer — the same text may consume roughly 35% more tokens, making long conversations hit the limit more frequently.

Issue 5: Slow Response / High Latency

claudeapi.com routes through optimized channels by default, with latency typically stable under 200ms. If you’re experiencing issues:

• Disable any proxy/VPN (claudeapi.com provides direct access — a proxy can actually slow things down)
• Check if your firewall is blocking gw.claudeapi.com
• Review real-time request logs in the claudeapi.com console to determine whether the issue is at the gateway or model level

11. FAQ

Q1: Are AutoClaw and OpenClaw the same thing?

No. OpenClaw and AutoClaw are different products. This tutorial is specifically for AutoClaw (the desktop Agent application with a local Gateway and Hermes Evolution Board).

Q2: Can I select OpenAI as the API Protocol?

No. AutoClaw’s Anthropic protocol path uses the native Messages API, which correctly handles Claude-specific fields like tool_use, cache_control, and thinking. Selecting OpenAI will cause you to lose these capabilities.

Q3: Can I add multiple Claude models at the same time?

Yes. Repeat Steps 3-5 with a different Model ID each time. Once added, the model selector at the bottom will show all your models — switch between them anytime with the dropdown.

Q4: Is Max Output 32000 enough?

For the vast majority of conversation scenarios, yes. If you’re doing long code generation or lengthy document translation, edit the model and increase it to 64000 (Sonnet 4.6 maximum) or 128000 (Opus 4.6 / 4.7 maximum).

12. Summary

Here’s the complete flow at a glance:

AutoClaw Client
    └─ Settings
        └─ Models & API
            └─ Add Custom Model
                ├─ Provider: Custom
                ├─ Protocol: Anthropic
                ├─ Base URL: https://gw.claudeapi.com
                ├─ Model ID: claude-opus-4-7
                ├─ API Key: sk-xxx (from claudeapi.com console)
                └─ Connectivity Test ✅ → Save → Set Default
                    └─ Back to main → Select model in bottom dropdown → Start chatting

AutoClaw Client
    └─ Settings
        └─ Models & API
            └─ Add Custom Model
                ├─ Provider: Custom
                ├─ Protocol: Anthropic
                ├─ Base URL: https://gw.claudeapi.com
                ├─ Model ID: claude-opus-4-7
                ├─ API Key: sk-xxx (from claudeapi.com console)
                └─ Connectivity Test ✅ → Save → Set Default
                    └─ Back to main → Select model in bottom dropdown → Start chatting

Once connected, every Agent inside AutoClaw (AutoClaw / DeepResearch / Cron / Browser Use / custom), every IM channel, and every Skill automatically shares the same Claude quota. Configure once, use everywhere.

👉 Sign up at claudeapi.com ｜ 💰 Go to Top-Up Center ｜ 🤝 Apply for Partnership

Screenshots and workflow in this tutorial are based on the AutoClaw release as of 2026-05-21. If the interface has been updated, defer to what you see in the app.