Choosing your agent's brain — what models are, what they cost, and how to set them up.
Your agent needs an AI "brain" to think with — that's the model. When your agent reads a message and writes a reply, it's the model doing the heavy lifting.
Different models have different strengths. Some are faster, some are smarter, some are cheaper. Think of it like choosing between a sports car, an SUV, and a commuter bike — they all get you there, but the experience (and the price) is different.
The good news: heyron lets you switch models anytime. You're never locked in. Try one, and if it doesn't fit, swap it in seconds.
Not to get started — your heyron subscription includes AI model usage. You can chat with your agent right away without any API key. However, if you want to use premium models heavily (like GPT-5.4 or Claude Opus), or if you run into usage limits, adding your own API key from the provider gives you more capacity. Think of it like this: the subscription gets you in the door, and your own API key is the upgrade if you need more horsepower.
Best for: General tasks, coding, creative writing
GPT-5.4 is the current flagship — performs on par with Sonnet and Opus for many tasks at a better price point. GPT-4o-mini is the budget option that still punches above its weight. GPT-5.3-codex is purpose-built for coding. Note: heavy use of premium models may require adding your own OpenAI API key.
Get API key → platform.openai.comBest for: Long documents, careful reasoning, following complex instructions
Claude models are known for being thorough and precise. If your agent handles long documents or nuanced requests, Claude shines.
Get API key → console.anthropic.comBest for: Research, analysis, multimodal (images + text)
Google's Gemini models handle text and images natively. Great value, especially the Flash tier for everyday use.
Get API key → aistudio.google.comBest for: Experimenting, accessing free/cheap models, variety
OpenRouter is a gateway to dozens of models from different providers. Great if you want to try things without committing to one provider. Some models are free or nearly free.
Get API key → openrouter.aiGo to the provider's website (links above) and sign up. Most offer free credits to start with.
Find the API keys section in your provider's dashboard and create a new key. It'll look something like sk-proj-abc123...xyz — a long string of characters.
Go to your heyron dashboard and configure the model in your agent's settings. Or just tell your agent:
"Set my OpenAI API key to sk-proj-..."
Never share your API key publicly. Never paste it in Discord, a public channel, or a GitHub repo. Treat it like a password. If you accidentally leak it, revoke it immediately on the provider's site and generate a new one.
Models charge per token — roughly 1 token ≈ 0.75 words. A typical back-and-forth conversation uses 1,000–5,000 tokens.
💬 Casual use (chatting, quick questions) — $5–20/month
💻 Heavy use (coding, long documents, frequent conversations) — $20–50/month
🧪 Experimenting (free models via OpenRouter) — $0
All major providers let you set a monthly spending cap so you don't get surprised. Do this before you start using the API.
OpenAI: Settings → Limits · Anthropic: Settings → Spend Limits · Google: Billing → Budgets
Start with a cheaper model like GPT-4o-mini or Gemini Flash. They're surprisingly good for most tasks. When you're ready to upgrade, GPT-5.4 is a great value sweet spot before jumping to Opus.
Switching is instant. You have two options:
Talk to your agent: Just say "Switch to Claude Sonnet" or "Use GPT-5.4". Your agent will handle the rest.
Use the dashboard: Go to your agent's settings and change the model in the Model section.
You can switch anytime — even mid-conversation. Different tasks might work better with different models, so don't be afraid to experiment.
Most model errors are account limits, missing provider credits, or key/provider mismatch. Use this checklist before opening a support ticket.
Add funds (or free credits) in your model provider account and confirm a spending limit is set above $0. If you're using your own API key, this error comes from the provider account attached to that key.
Regenerate the key in the provider dashboard, then re-add it in heyron. Watch for copy/paste issues (leading/trailing spaces are common). Make sure the key matches the provider (OpenAI key with OpenAI model, Anthropic key with Claude, etc.).
This is usually stale state (old session/runtime still holding old auth). Save the new key, then start a fresh chat/session and run one tiny test prompt. If you recently regenerated a key, make sure any old key is removed/disabled so requests can’t silently hit the wrong credential.
If behavior still looks identical, switch to one known-good model for that same provider and ask the agent to quote the exact provider error text before continuing.
You likely hit rate limits or monthly caps. Either lower traffic, switch to a cheaper/faster model for routine tasks, or raise your provider spend/rate limits. Keep a fallback model configured so your agent stays responsive.
For most users: no, not at first. heyron subscription usage covers getting started. Add your own key when you need higher-volume premium usage, custom provider routing, or you're hitting included-usage limits.
This means the provider key hit a hard limit (daily/monthly cap, total spend cap, or exhausted balance). Open the provider key dashboard, confirm remaining credits, and check per-key limits. Then either top up, raise the cap, or switch to another key/model you already have access to.
If the error appears after a model switch, verify that the selected model is actually available to that exact key.
This usually means the model string is malformed or mapped to the wrong provider namespace. Use exact provider/model IDs (for example openrouter/minimax/minimax-m2.5 only if your route actually uses OpenRouter). Don’t combine one provider name with another provider's model slug.
Fast fix: switch to one known-good model first, confirm it works, then change only one field at a time (primary model, then fallback). If primary + fallback both fail immediately, your IDs or provider route are mismatched.
Subscription credits cover a lot, but some external providers (image/video/voice APIs) can have their own billing rules, quotas, or premium tiers. If an external provider returns billing/limit errors, that limit is outside core chat usage and must be handled in that provider account.
Quick check: ask the agent to quote the exact provider error text. If it mentions quota/credits/rate limit from a third-party service, troubleshoot that provider first.
This is usually a context/session issue, not a broken model switch. Start a fresh chat/session, then ask the agent to confirm the active model in one line (for example: "Confirm active model only"). If needed, switch once via dashboard settings, then run one clean test prompt. Old context can make two different models sound similar on routine tasks.
There are two different buckets people mix up: (1) included heyron usage, and (2) your own provider key usage. Provider-key limits reset on that provider's schedule (daily/monthly depending on account settings), not on heyron's internal cycle.
If you're unsure which limit you hit, ask the agent to quote the exact error text and provider name. If the error says OpenRouter/OpenAI/Anthropic quota or key-limit, fix it in that provider dashboard first.
Some external tools still require manual confirmation/selection steps in their own UI. Your agent can draft prompts and workflows, but image picking/upscaling may need a human click unless you use an API-native pipeline. This is expected behavior, not a model failure.