Pay per page. Use as much as you want.

Sign up for 100,000 starter credits. Pay just 100 credits per page of agent compute.

Start with $10 in credits → See the pricing model ↓

How pricing works.

You pay per page of agent compute. A page is 3,000 characters — about a screen of text, or roughly 600 words. Agent compute covers anything your agent does on your behalf — chat messages, background tasks, scheduled events.

Configuration Rate When it's used
Standard 100 credits per page Everyday agents: coaching, Q&A, support, research, content.
Standard-TEE 200 credits per page Confidential workloads: legal, financial. Hardware-isolated compute. Even we can't see what's being processed.
Confidential compute available.
Standard-TEE runs inside a hardware-isolated Trusted Execution Environment — even we can't see what your agent processes. The agent's author picks Standard or Standard-TEE at publish time. See how it works →

Chat conversations have a 2-page minimum.

Free to start. 100,000 credits on signup.

When you sign up, you get 100,000 BlueNexus credits — worth $10. Enough for hundreds of short questions or a handful of deep multi-turn conversations before you decide to top up. No card needed at signup.

Sign up free →

Worked example.

A 5-message coaching conversation at ~4,500 chars of new content per turn. Each turn bills against the running conversation:

Turn 1: 2.0 pages × 100 cr = 200 cr
Turn 2: 3.0 pages × 100 cr = 300 cr
Turn 3: 4.5 pages × 100 cr = 450 cr
Turn 4: 6.0 pages × 100 cr = 600 cr
Turn 5: 7.5 pages × 100 cr = 750 cr
Total: 2,300 credits on Standard (4,600 on Standard-TEE).

Short Q&A (one turn, 2-page minimum): 200 credits. Long deep multi-turn discussions can run into 5,000–10,000+ credits as the running conversation grows.

You pay for what the agent actually processes. Each turn, the agent re-reads the full context — we charge in proportion to that work. Short conversations cost very little; deep ones cost more.

What counts toward a page.

Each turn bills against the running conversation — the agent re-reads the full context each turn, and we charge for what it processes. Here's the full audit:

Counts

  • Your messages
  • The agent's replies
  • The agent's reasoning (when reasoning is on)
  • Tool results that produce new data
  • Tool calls that generate content (new documents, memory, image prompts etc.)
  • Knowledge-base + memory content the agent fetches for you

Doesn't count

  • The agent's system prompt and personality
  • Tool definitions sent to the model on every call
  • Hidden internal state the agent never shows you
  • Platform maintenance behind the scenes — catalogue indexing, health checks, monitoring

If the agent processed it on your behalf, it's part of what you pay for. If it's platform plumbing, we absorb it.

Image generation is charged separately.

Agents that generate images charge a separate per-image fee, priced by fidelity (Draft, Standard, Premium) and size (square, landscape/portrait, 2K, 4K).

Fidelity Square Landscape / Portrait 2K 4K
Draft 150 cr 200 cr 400 cr 800 cr
Standard 400 cr 500 cr 1,000 cr 2,000 cr
Premium 800 cr 1,200 cr 2,000 cr 4,000 cr

Most common: a standard square image is 400 credits. If your conversation doesn't trigger images, you don't pay for image capacity — images aren't generated unless your agent does so as part of its response.

Note: image generation runs on standard infrastructure, not in confidential compute — even on Standard-TEE agents.

Common questions.

Do longer conversations cost more?
Yes. Each turn bills against the running conversation × the per-page rate. As your chat grows, each new turn covers more pages — the agent re-reads the full context each turn, and we charge for what it actually processes. Short conversations cost very little; long deep conversations cost more, proportionally. You pay in proportion to the agent's actual work, not just the new content added.
What counts toward a page?
Counts: your messages, the agent's replies, the agent's reasoning when it's on, tool results that produce new data, tool calls that generate content (new documents, memory, image prompts etc.), and knowledge-base or memory content the agent fetches in response to you. Doesn't count: the agent's system prompt and personality, tool definitions sent to the model, hidden internal state, platform maintenance behind the scenes (indexing, health checks). If the agent processed it on your behalf, it's part of what you pay for. If it's platform plumbing, we absorb it.
Why is there a 2-page minimum per conversation?
Because even a one-word "yes" requires the agent to read the full context to know what it's responding to. The 2-page minimum (200 credits Standard / 400 credits Standard-TEE) covers the real cost of processing the conversation context, no matter how short your first message. Above the minimum, billing is exactly proportional.
Why are there two configurations?
Some agents need confidential compute. Standard runs on regular infrastructure at 100 credits per page. Standard-TEE runs inside a hardware-isolated Trusted Execution Environment — even we can't see what's being processed — at 200 credits per page. The agent's author picks; you can upgrade to Standard-TEE on a specific conversation if you want extra confidentiality, but you can't downgrade below the author's floor.
What's a credit worth?
1 credit = $0.0001. 10,000 credits = $1. The $10 starter grant gives you 100,000 credits — enough for hundreds of short questions or a handful of deep multi-turn conversations.
What about image generation?
Agents that generate images add a per-image fee on top of the per-page rate, priced by fidelity (Draft, Standard, Premium) and size (square, landscape/portrait, 2K, 4K). A typical standard square image is 400 credits; a draft square is 150 credits; a premium 4K image is 4,000 credits. The full matrix is shown above and on each agent's listing where image generation is enabled. If your conversation doesn't trigger images, you don't pay for image capacity.
Can I just use the starter credits forever?
Until you run out. 100,000 credits buys hundreds of short questions or a handful of long multi-turn conversations. Short tries use very little; deep conversations cost more, proportionally. You'll never auto-renew into a paid plan; you top up when you're ready.
What about model selection? Which model is actually running my agent?
We pick the best model available at each configuration's price point and route accordingly. The configuration label (Standard / Standard-TEE) is our commitment to capability + confidentiality at the published rate — we don't lock to specific vendors so we can keep upgrading what runs underneath.
What about SOC 2 / compliance paperwork?
SOC 2 audit is underway — current status, security overview, and questionnaire responses are available on request via /security/ or /contact/. Sensitive workloads are well-served today by our confidential-compute option (Standard-TEE).

Not buying for yourself?

Different shapes of BlueNexus have different pricing conversations.

Publishing + earning

Earn each time your agent is used. For experts →

Embedding in your product

Per-end-user pricing fitted to your model. For products →

Start with $10 in credits.

No subscription. No card needed to start. The grant gives you room to build your first agent and try a handful of others before you decide to top up.

Start free → Talk to us →

Subscribe to our newsletter

Get product insights, AI and product thinking delivered to your inbox.