Pick any model. Bring your own key. Go open source. Pay only processing costs. We host everything — you decide what runs under the hood.
The honest bit
Bring your own key.
Add your own API key for any supported provider and pay them directly — at their published rates. We never mark up third-party usage.
Open source? Yes.
DeepSeek R1, Llama 3.3, and Mistral Small are MIT or Apache 2.0 licensed. Strong models, fraction of the cost, no proprietary lock-in.
We host everything.
No infrastructure to manage. No API wrappers to build. Plug in your key, pick your model, and it works inside your Aigent immediately.
No guarantees. Ever.
No AI model — not Anthropic, not Google, not us — guarantees accuracy. Results vary. For property, compliance, and legal matters, always verify with a qualified professional.
Cheaper = smarter choice.
For routine tasks — booking confirmations, maintenance acknowledgements, FAQ replies — a cheaper model works just as well. Save the powerful models for negotiations and complex queries.
Legal tier is fixed.
Eviction assessments, deposit disputes, and contract reviews always use our Opus instance — this is a governance rule, not configurable. No cost or key arrangement changes that.
Included with your plan
These run on our Anthropic account. No key needed — processing costs are billed to your subscription.
Sonnet 4.6
$3 in · $15 out / 1M tokens
Balanced reasoning, negotiations, complex queries. Best for most use cases.
Haiku 4.5
$0.80 in · $4 out / 1M tokens
Fast and affordable. Ideal for routine tasks, FAQs, and simple confirmations.
Bring your own key
Add your own API key in Models → Your API keys inside the platform. You pay the provider at their published rate. We charge nothing on top.
Keys are encrypted at rest and never returned to the client after saving.
Three open source options available
DeepSeek R1, Llama 3.3 70B, and Mistral Small are MIT or Apache 2.0 licensed. Comparable quality. Significantly lower cost.
DeepSeek R1
$0.55 in · $2.19 out / 1M
MIT-licensed reasoning model. Open source, strong performance at low cost.
via DeepSeek
Llama 3.3 70B
$0.18 in · $0.40 out / 1M
Apache 2.0. Meta's open-source flagship. Extremely affordable at scale.
via Meta
Mistral Small
$0.10 in · $0.30 out / 1M
Apache 2.0. Efficient multilingual model. One of the cheapest options available.
via Mistral
Gemini 2.0 Flash
$0.10 in · $0.40 out / 1M
Fast multimodal model from Google. Well-suited for high-volume processing.
via Google
GPT-4o Mini
$0.15 in · $0.60 out / 1M
OpenAI's affordable option. Familiar outputs, reasonable pricing.
via OpenAI
Grok 3 Mini
$0.30 in · $0.50 out / 1M
xAI's fast reasoning model. Real-time data access on the full Grok 3.
via xAI
How it works
01
Get an API key
Sign up with any supported provider — Anthropic, Google, OpenAI, xAI, Mistral, Meta, DeepSeek. Generate an API key from their dashboard.
02
Add it to Raia
Open your Aigent panel → Models → Your API keys. Paste your key. It's encrypted with AES-256-GCM immediately and never shown again.
03
Pick your model
Select any model unlocked by your key. Your Aigent will use it from that point on. Swap anytime. Remove your key anytime.
Cost comparison
A typical tenant enquiry — booking a viewing — runs about 400 input tokens and 80 output tokens. Here's what that costs on each model.
| Model | Provider | Per enquiry | 1,000 enquiries |
|---|---|---|---|
| Mistral SmallOS | Mistral | <$0.0001 | £0.05 |
| Gemini 2.0 Flash | <$0.0001 | £0.06 | |
| Llama 3.3 70BOS | Meta | $0.00010 | £0.08 |
| GPT-4o Mini | OpenAI | $0.00011 | £0.09 |
| DeepSeek R1OS | DeepSeek | $0.00040 | £0.32 |
| Haiku 4.5 | Anthropic | $0.00064 | £0.51 |
| Sonnet 4.6 | Anthropic | $0.00240 | £1.92 |
Assumes 400 input + 80 output tokens per enquiry. Exchange rate $1 = £0.80. OS = open source. Actual costs vary with message length. BYOK models billed directly by your provider.
Questions
No. Platform models (Haiku and Sonnet) are always available on our Anthropic account — no key needed. BYOK is optional, for users who want a different model or want to control their own spending.
Yes. Open your Aigent panel, go to Models, pick a different model, and save. The change takes effect immediately on the next message. You can switch back anytime.
Keys are encrypted with AES-256-GCM immediately on receipt, stored encrypted in the database, and never returned to the client. They are only decrypted server-side at inference time and scoped to your account only.
That's the tradeoff. Cheaper models are less capable at reasoning and nuance. We recommend Sonnet for anything involving judgment — negotiations, complaints, compliance queries. Haiku or budget third-party models for routine tasks. No model is guaranteed to be correct.
Opus is reserved for legal-grade tasks — eviction assessments, deposit disputes, contract reviews — which require CEO/CTO approval before sending. This is a governance requirement. Opus for general chat provides no benefit over Sonnet and costs ~5× more.
DeepSeek R1 (MIT), Llama 3.3 70B (Apache 2.0), and Mistral Small (Apache 2.0) are models whose weights are publicly available and freely usable. When you use them via BYOK, you pay the respective provider for compute — not for the model weights.
No model gives a guarantee.
AI outputs — from any provider, at any price point — are not legal, regulatory, financial, or professional advice. Results vary. Models can be wrong, outdated, or misleading. Always verify anything material with a qualified professional before acting. This applies to Raia, and to every third-party model available through BYOK.
See real processing costs.
Token counts and pound estimates for real property conversations.