Explore proposed plans and workload costs while Core is in private preview. Confirm access and commercial terms with the team. Your model usage stays on your provider’s bill.
From your first agent to your next release.
Proposed prices in USD, before tax.
A 12-month commitment, billed up front. Extra credits, storage and transfer are billed each month.
Developers evaluating the integration
Request accessCore capabilities, the sample integration, capped development usage, short retention and self-service support.
A small team releasing its first embedded workflow
Request accessPlanned production tier with embedded UI, workflow and knowledge capabilities, modest capacity and standard support.
A software business using the platform in customer workflows
Request accessMore concurrency and usage, more projects, longer history, team controls and priority support.
A company with substantial or growing production usage
Request accessLarger capacity, better usage economics, longer retention, broader organization controls and defined support.
Buyers with specific deployment or contract needs
Talk to usA scoped agreement for hosting, capacity, retention, support and procurement.
Illustrative workloads using the proposed price list and your annual billing selection. Each separates the subscription, additional platform consumption and model-provider charges. These examples use capabilities implemented for private preview; prices and performance are planning assumptions.
1,000 submissions each month, checked against a small knowledge collection. One run and two event deliveries per submission.
1,000 runs × 30 active seconds ÷ 3,600 = 8.33 active run-hours.
Launch: up to 10 concurrent runs. Stored documents exceed Sandbox’s 100 MB allowance.
Subscription shown as a monthly equivalent; $1,188.00 paid annually. Additional platform usage billed monthly. USD, before tax.
Storage is the monthly average retained; indexing is new or changed source text during the month. Delivery counts include any retries.
Per run, across all model calls: 2,500 input tokens and 500 output tokens.
Embedding volume assumes 250 million tokens per GB of indexed source text. Actual token counts depend on your content and tokenizer. Model tokens do not consume Core credits.
500 documents each month. Each workflow drafts, waits four hours for a reviewer, then makes one revision. The 180 active seconds cover drafting and revision; review waiting time is excluded.
500 runs × 180 active seconds ÷ 3,600 = 25 active run-hours.
Launch: assume no more than 10 concurrent runs, with 20 GB of retained files and four event deliveries per workflow.
Subscription shown as a monthly equivalent; $1,188.00 paid annually. Additional platform usage billed monthly. USD, before tax.
Storage is the monthly average retained; indexing is new or changed source text during the month. Delivery counts include any retries.
Per run, across all model calls: 12,000 input tokens and 3,000 output tokens.
Embedding volume assumes 250 million tokens per GB of indexed source text. Actual token counts depend on your content and tokenizer. Model tokens do not consume Core credits.
25 customer organizations × 2,000 tasks each month = 50,000 runs. A shared project embeds the feature in your SaaS, with two event deliveries per task.
50,000 runs × 90 active seconds ÷ 3,600 = 1,250 active run-hours.
Growth: assume up to 50 concurrent runs. Your application supplies business-data storage and authorization. Lower concurrency may allow a cheaper plan.
Subscription shown as a monthly equivalent; $5,988.00 paid annually. Additional platform usage billed monthly. USD, before tax.
Storage is the monthly average retained; indexing is new or changed source text during the month. Delivery counts include any retries.
Per run, across all model calls: 6,000 input tokens and 1,000 output tokens.
Embedding volume assumes 250 million tokens per GB of indexed source text. Actual token counts depend on your content and tokenizer. Model tokens do not consume Core credits.
Enter what you expect to use. Start with current private-preview capabilities, or include planned capabilities to explore a future workload. Plan fees follow the billing chosen above: now annual.
Current capabilities only: planned meters are hidden and excluded from the total. Current means implemented for private preview; access is by request.
Launch · Billed annually · Current capabilities only
Plus $2.15 of storage and transfer on paid plans.
Lowest cost for this usage and concurrency: Launch at $102.65 a month, billed annually, including 1.5 extra credits.
Prices in US dollars, before tax, including storage and transfer. Usage and concurrent runs are included in this calculation; check project, retention and other plan limits too. Your AI provider's charges are separate.
Your plan includes monthly credits for compute. Add more at your plan’s rate, with storage and transfer billed separately. Model tokens go directly to your provider.
You connect your own AI provider and it bills you directly. We show token use as an estimate and never mark it up.
Set a monthly cap with alerts. At the cap, new work waits for you. Work already running finishes.
Extra credits are billed monthly within your cap. Scale and Enterprise can prepay credits at their rate, usable for 12 months.
Usage is counted from each operation’s record, so a recovered step never counts twice. Storage and data sent out use flat rates.
| What uses credits | Unit | Credits |
|---|---|---|
| Run timeTime a run is working: thinking, calling tools, writing. Waiting for a person, an outside signal or a timer costs nothing. | active run-hour | 0.35 |
| Sandbox compute, smallPlannedCode that store items such as the Node interpreter run for your agents, billed by the second for its size. | hour (1 vCPU, 2 GB) | 0.18 |
| Sandbox compute, mediumPlannedThe same, at twice the size. | hour (2 vCPU, 4 GB) | 0.36 |
| Sandbox compute, largePlannedThe same, at four times the size. | hour (4 vCPU, 8 GB) | 0.72 |
| Knowledge indexingReading, splitting and indexing documents into a collection. Embeddings run on your own provider. | GB of source text | 5.5 |
| Event deliveriesSigned webhooks and SQS messages about runs, including retries we send for you. | 10,000 deliveries | 0.5 |
| Channel messagesPlannedMessages in and out through Slack, Teams, Discord and WhatsApp installations. | 1,000 messages | 1 |
| Model tokensPaid to your provider. Shown as an estimate. | token | 0 |
| Storage and transfer, at a flat rate | Unit | Price |
|---|---|---|
| File storageWorkspace files, delivered results and the source documents of knowledge collections, with their versions. | GB-month | $0.08 |
| Knowledge index storageThe searchable index of a knowledge collection: its passages and their embeddings, with backups. | GB-month | $0.60 |
| Database storagePlannedShared databases that agents keep across runs, with point-in-time backups, per database or per audience partition. | GB-month | $0.50 |
| Data transfer outResults, files and streamed answers sent out of the platform to your servers and your users' pages. | GB | $0.15 |
Compare every limit by plan. Limits are checked when work starts, never in the middle of a run.
Scroll the table to compare all plans on smaller screens.| Limit | SandboxFree | Launch$99/mo | Growth$499/mo | Scale$1,499/mo | EnterpriseCustom |
|---|---|---|---|---|---|
| Usage | |||||
| Credits included each month | 10 | 75 | 450 | 1,500 | Contracted |
| Extra credits | Not available | $1.00 each | $0.90 each | $0.75 each | Contracted rate |
| Runs at a time | 2 | 10 | 50 | 250 | Custom |
| Longest runOn paid plans, you set your own limits per agent or workflow. | 10 minutes | No limit | No limit | No limit | No limit |
| API requests per minute | 60 | 600 | 3,000 | 10,000 | Custom |
| Stored data | 100 MB | Flat rate, no cap | Flat rate, no cap | Flat rate, no cap | Custom |
| Data transfer out | 1 GB a month | Flat rate, no cap | Flat rate, no cap | Flat rate, no cap | Custom |
| Workspace | |||||
| Projects | 1 | 3 | 10 | 50 | Custom |
| Environments per project | Development only | 2 | 5 | Unlimited | Unlimited |
| Members | 3 | 5 | 25 | 100 | Custom |
| Knowledge collectionsTheir size is billed too, as indexing credits and index storage. | 2 | 5 | 8 | 20 | Contracted |
| Retention | |||||
| Run history, events and logsCounted from when a run finishes. A run still in progress or waiting is kept. | 7 days | 30 days | 90 days | 1 year | Custom |
| Run files and delivered resultsDownload them, or save them to a shared workspace, to keep them longer. | 7 days | 30 days | 90 days | 1 year | Custom |
| Audit history | 30 days | 90 days | 1 year | 3 years | Custom |
| Recovery after you delete something | — | 7 days | 14 days | 30 days | Custom |
| Knowledge, shared workspaces and databasesBilled as storage at the flat rate. Sandbox keeps up to 100 MB. | Until you delete them | Until you delete them | Until you delete them | Until you delete them | Until you delete them |
| Embedding and integration | |||||
| Websites that embed your agents | localhost only | 2 | 10 | Unlimited | Unlimited |
| Embedded users | 10 test users | Unlimited | Unlimited | Unlimited | Unlimited |
| Webhooks and SQS destinations | 2 | 10 | 50 | Unlimited | Unlimited |
| Shared databases (planned) | — | 3 | 20 | 100 | Custom |
| Channel installations (planned) | — | 1 | 5 | 25 | Custom |
| Usage by audience, for rebilling | — | Yes | Yes | Yes | Yes |
| Controls and support | |||||
| Roles, approvals and audit history | Yes | Yes | Yes | Yes | Yes |
| Monthly spend cap with alerts | — | Yes | Yes | Yes | Yes |
| Audit export | — | — | Yes | Yes | Yes |
| Single sign-on and SCIM | — | — | — | Yes | Yes |
| Dedicated hosting and region choice | — | — | — | — | Yes |
| Support | Docs | Email, 2 business days | Priority, 1 business day | Named contact, agreed response times | Contracted |
Plan fees, credit billing, and commercial limits below describe the proposed commercial model. Confirm terms before purchase.
Access is arranged by invitation. Public signup and public npm packages are not available yet. Contact the team to confirm an environment, API credentials, package access, and the capabilities you can evaluate. Request preview access →
Annual is a 12-month commitment with the plan fee billed up front, and costs 17% less. Monthly can stop at the end of any month. Either way, extra credits, storage and data out are billed each month at the same rates, and the included credits renew monthly. You can move from monthly to annual at any time. Enterprise is billed annually.
No. A run waiting for a person, an outside signal or a timer holds no worker, so it uses no run time. Its stored files still count as storage.
New work is refused when it starts, with a message that says which limit. Work already running finishes. On Sandbox, work pauses until next month once the 10 credits are used. On paid plans, extra credits are billed until your spend cap.
Runs, their events and their files are deleted once they're older than your plan keeps them, counted from when each run finished. Knowledge collections, shared workspaces and databases stay until you delete them. Moving to a bigger plan keeps what's still there for longer. Moving to a smaller one applies its shorter retention from then on.
Credits pay for compute, which is where the platform does work for you. Storage and data sent out are billed at a flat rate close to what they cost, so keeping knowledge, files and databases with us stays cheap as they grow. Sandbox includes 100 MB of storage and 1 GB a month of data out.
Credits included in a plan are for that month. Prepaid credits on Scale and Enterprise last 12 months from purchase.
Current reports show estimated AI usage by audience. Platform credit accounting and invoicing are planned. Your application remains responsible for choosing its prices and billing its own customers.
Any provider your environment connects, with your own key. Agents can fall back between models. Token use is shown as an estimate, and the provider's bill is the real one.
Explore the app experience, follow the workflow, and find the right place to begin.