Choose the right model for the job
- Use lighter or cheaper models for repeated drafts, summaries, and classification work.
- Move to heavier models when the task needs deeper reasoning, larger refactors, or more complex planning.
- Save your most-used model slugs locally so you are not retyping them from memory.
Control context size
Long-context work is powerful, but it becomes expensive and harder to debug when every request carries too much irrelevant material. Clean inputs usually outperform oversized ones.
- Send only the files, logs, or text that matter for the current task.
- Summarize earlier steps before you continue instead of pasting entire history chains forever.
- Break large workflows into stages: inspect, decide, implement, verify.
Prompt more specifically
- Tell the model what output format you want.
- State the constraints that matter, especially if a tool or integration has non-obvious rules.
- Give the model one clear objective before adding optional extras.
Reuse the prompts that work
When you find a system prompt, agent instruction block, or code-review template that produces clean results, keep it in your local setup notes or tool settings. Stable prompt fragments save time and reduce noisy retries.
Protect your access
- Keep the key in local environment variables, app secrets, or a real secret manager.
- Do not paste real keys into public issue trackers, forum posts, or screenshots.
- Claim your order in the portal if you want a safer long-term place to review access and order history.
Use the portal and docs together
The portal is useful for order history, delivery visibility, and usage-related account information. The docs are better for setup logic, model references, and troubleshooting. Using both reduces avoidable support friction.