300+ models, one API key
Claude, GPT-5, Gemini 2.5 Flash, Llama 4, Mistral, Qwen, Command R+, and every model OpenRouter adds. One field in your agent markdown. One key. No separate accounts per provider.
Model-agnostic AI platform
Pick the model. Swap it in one line.
300+ models. Haiku for triage. Sonnet for daily work. Opus for one-shot analysis. Change one field in the agent markdown and the platform routes accordingly.
No vendor lock-in · BYOK supported · Free plan, no card needed
Agent markdown: model: anthropic/claude-haiku-4-5 | skills: strategy-review / model: anthropic/claude-opus-4-7. Terminal: oneie agent publish ✓
300+
Models available through one API key
95%
Price drop for comparable quality — 18 months
<150ms
First token on Groq hot paths
1
Fields to change for a complete model migration
"The companies that win this decade won't be the ones with the smartest models. They'll be the ones with the shortest distance between a customer's question and a useful answer."
How model routing works
Pick, bind, fallback, swap. In that order.
1
Pick a model
Set the model: field in your agent frontmatter. anthropic/claude-haiku-4-5 for high-volume triage, claude-sonnet-4-7 for daily agency work, claude-opus-4-7 for deep one-shot analysis.
model: {provider}/{model-id} · OpenRouter catalogue · instant
2
Set per-skill overrides
Individual skills override the agent default. Triage on Haiku, proposals on Opus. One conversation using two models costs what the tasks actually warrant, not what the most expensive model in the fleet costs.
skills: - name: qualify / model: haiku · skills: - name: proposal / model: opus
3
The fallback chain covers you
OpenRouter gateway first. Groq direct when speed matters. AI SDK Gateway as fallback. Degraded mode (cached or structured error) when nothing is reachable. You do not configure this. It is built in.
OpenRouter → Groq → AI SDK Gateway → degraded · no configuration
4
Swap when better arrives
Change one string in the agent file. Run the eval gate. Roll to clients. The tool definitions, the instructions, the substrate wiring — all unchanged. The model is the only thing that moves.
1-field change · eval gate · roll to clients · no rebuild
What model-agnostic means in practice
One key. Every model. Your margin.
Per-skill model binding
Set the default at the agent level. Override per skill. Draft posts on Haiku at 25 credits/1k tokens. Strategy review on Opus at 1,500. One conversation, two models. The cost tuning is at the task level, not the agent level.
Groq for speed
Sub-200ms first token on Groq-served hot paths. Custom inference silicon. For interactive tasks where the user is waiting and the decision is not complex, latency is the product.
Four-link fallback chain
OpenRouter gateway → Groq direct → AI SDK Gateway → degraded mode. A provider having a bad hour does not silence your agents. The fallback chain is built in. Your agents keep running.
BYOK — bring your own key
Existing enterprise agreement with Anthropic, OpenAI, or Google? Bring your own key. Agents that target that provider route through your key, not the platform's pool. Same routing logic. Your provider bill.
Margin grows as costs fall
Claude Opus launched at $15/M input tokens. 16 months later, Haiku at comparable quality costs $0.80/M. A 95% reduction. When model costs fall and you hold client rates, the margin expansion is automatic. No renegotiation.
Model flexibility vs. the alternatives
| Feature | ONE | Single-model SaaS tool | Build your own router |
|---|---|---|---|
| Model flexibility (pick any, swap freely) | |||
| 300+ model catalogue via one key | |||
| Per-skill model binding | |||
| Pre-built fallback chain | |||
| BYOK (enterprise agreements) | |||
| Margin grows as costs fall automatically | |||
| Total control of routing logicHonesty: building your own router wins here — costs 4–6 weeks engineering time + ongoing maintenance. |
Same model catalogue on every plan
Credits are the unit. The model is your choice. Annual billing pre-selected.
Starter
All models · standard routing
$500/mo
5M credits/mo
Full 300+ model catalogue
Groq fast-path access
Per-agent model binding
Email support
Most popular
Agency
Per-skill binding · BYOK · daily caps
$5,000/mo
60M credits/mo
Per-skill model binding
BYOK for enterprise agreements
Daily credit caps per agent
Priority edge routing
Priority support
Scale
Reserved capacity · SLA 99.99%
$50,000
750M credits/mo
Reserved Groq capacity
Dedicated model routing tier
Custom fallback chain configuration
SLA 99.99%
Dedicated success manager
Questions about models
The model is not the moat. The routing is.
300+ models. One field to switch. The fallback chain keeps running. Your margin grows when costs fall.
No vendor lock-in · Free plan, no card needed