AI Model Configuration Guide
OpenClacky is an AI Coding Agent tool — it doesn't come with a built-in "brain". You need to configure an AI model (via API Key) so it can write code, debug, and edit docs for you.
This doc explains in plain terms: what a model is, why you configure it yourself, and which one to pick. After reading it, you'll know exactly what to do next.
Agent vs. Model: The One-Minute Explanation
Here's an analogy:
| Role | Think of it as | What it does |
|---|---|---|
| OpenClacky (Agent) | A capable assistant | Types, operates your machine, calls tools, takes notes |
| AI Model (e.g. DeepSeek, GLM) | The assistant's "brain" | Understands your requests, thinks, generates answers |
| API Key | The "admission ticket" to the brain | Lets the Agent call the model; you pay per usage |
No model = an assistant with hands and legs but no brain. Useless.
💡 Agent is the shell, model is the chip, API Key is the key.
Why Doesn't OpenClacky Ship a Built-in Model?
First-time users often ask: "Why do I have to go buy a Key myself? Can't you just include one?"
This is the industry standard, for 3 reasons:
1. Model calls cost money, charged per usage
Every conversation consumes compute, and model providers charge per token (like a utility meter). If OpenClacky paid the bill for everyone, we'd go bankrupt the moment we got popular.
2. You should have full control
- Which model, how much you spend, when to stop — all up to you
- Your data goes directly to the provider you choose; OpenClacky doesn't intercept it
- Want to switch providers? Change a config, no lock-in.
3. Different models have different strengths
One is best for coding, another for Chinese understanding, a third for price-per-token. Giving you the choice means you can pick the right tool for each task.
Which One Should I Pick?
We currently recommend these 4 providers — each suited to different use cases:
⭐ OpenClacky AI Keys — Recommended First Choice
Best for: users who want top-tier agent performance without the hassle
If you want the best models, don't want to juggle multiple accounts, and don't want to research which model performs best for which task — OpenClacky AI Keys is built for you.
One Key, access to the top model lineup:
OpenClacky AI Keys aggregates the most capable models available today, including:
- Claude series (Anthropic): widely regarded as the best model for Coding Agents; top-tier instruction following and code generation
- Gemini series (Google): standout multimodal capabilities; great at long-document understanding and analysis
- DeepSeek V4 (DeepSeek): high-value flagship; exceptional logic reasoning and coding at a fraction of the cost
These models typically require separate account registrations and juggling multiple billing setups — high friction. OpenClacky AI Keys removes all of that: one Key, one top-up, switch models anytime.
Built-in caching reduces your real cost:
OpenClacky applies prompt cache-hit optimization under the hood. For the repetitive system prompts common in Agent tasks, cache hits significantly reduce token costs — your actual spend is well below what you'd pay hitting the official APIs directly.
DeepSeek — Long Context + Best Value
Best for: users who need large context windows and want to keep costs low
DeepSeek V4 supports a 1 million token context window — one of the largest among mainstream models — while keeping token prices highly competitive. It excels at code generation, logical reasoning, and math, with significantly improved Agent capabilities. A strong choice when you need "large context + strong reasoning + low cost" all at once.
Good for: long-document understanding and analysis, code completion and refactoring, data analysis scripts, Agent automation tasks.
GLM (Zhipu AI) — Long-Horizon Autonomous Execution
Best for: users who need AI to autonomously complete complex engineering tasks end-to-end
GLM-5.1's standout capability is long-horizon task execution — it can work autonomously for up to 8 hours in a single session, completing the full loop from planning and execution to testing and delivery. In Agentic Coding scenarios, it handles multi-stage tasks with strong dependencies particularly well. It also brings deep strengths in tool calling, multi-turn dialogue, and Chinese semantic understanding.
Good for: long-horizon Coding Agent tasks, iterative engineering projects, complex tool-calling workflows, Chinese content creation.
MiniMax — Agent Workflows & Engineering Delivery
Best for: users building complex Agent pipelines or handling multi-step engineering delivery
MiniMax M2.7 is purpose-built for Agent workflows and real-world software engineering. It can autonomously build Agent Harnesses, orchestrate complex tool chains, and execute stably in multi-agent collaboration (Agent Teams) scenarios. It also performs well in professional office tasks (complex Excel/PPT/Word editing) and domain-specific tasks like financial analysis.
Good for: end-to-end software engineering projects, multi-agent collaboration, professional document processing, complex Agent workflow construction.
Quick Decision Guide
| My situation | Recommended |
|---|---|
| Want Claude / Gemini without registration hassle | OpenClacky AI Keys |
| Need long context window + cost efficiency | DeepSeek |
| Need long-horizon autonomous task execution | GLM |
| Building Agent pipelines or end-to-end engineering delivery | MiniMax |
How Do I Use the Key After I Get It?
On first launch, the OpenClacky client opens an Onboarding flow. The second step — "Connect AI Model" — is where you plug the Key in. You only need to:
- Choose Provider: Pick the one you just bought a Key from (OpenClacky / DeepSeek / GLM / MiniMax, etc.)
- Choose Model: The dropdown auto-populates with available models for that provider — pick one
- Confirm Base URL: Auto-filled once you pick a provider; you normally don't need to change it
- Paste the API Key: Paste the Key you copied from the provider's dashboard
- Click "Test & Continue" — once it passes, you're in
💡 That's it — Provider and Base URL are built-in, you really just "pick one + paste one".
💡 Want to switch models or add more Keys later? Edit in Settings anytime, no reinstall needed.
Security Reminders
An API Key is like a credit card — if it leaks, others can burn through your balance. Rules:
- ❌ Never paste API Keys in group chats, commit to GitHub, or share in screenshots
- ✅ Do set spending limits (most providers support this)
- ✅ Do check usage regularly; if anything looks off, rotate the Key immediately
- ✅ Do use separate Keys per project/device so you can trace usage
Next Step
Pick a provider and follow the matching guide: