Plans under evaluation
Free to run yourself. Hosted plans are proposed, not on sale.
Open source at the core, hosted in the cloud when you want it. You bring your own model keys on every tier, so inference is billed by your provider and never sits inside a plan. Community, PRO and Max are proposed tiers. Hosted pricing, quotas and availability remain subject to validation before a paid launch. No plan promises unlimited compute.
Community
Coming soonFree
Self-hosted. Not released yet: the repository is private and no package is published
The local engine and CLI on your own machine.
- Local engine and CLI
- Mock provider and benchmark generator
- HTML and JSON reports
- Your own provider account
Inference: billed by your own provider account. Not included.
Read the quickstartPRO
Proposed, not on sale$29
per month, excluding tax
Price and included usage are not finalized.
One hosted workspace for a small team.
- 1 hosted workspace, up to 3 members
- 2 concurrent runs
- 8 model calls in flight
- 30-day evidence retention
Inference: billed by your own provider account. Not included.
Get early accessMax
Proposed, not on sale$99
per month, excluding tax
Price and included usage are not finalized.
Several workspaces and managed policy canaries.
- Up to 5 workspaces, 10 members
- 5 concurrent runs
- 32 calls in flight per workspace
- 90-day retention, policy canaries
Inference: billed by your own provider account. Not included.
Get early accessCompare the proposed tiers
Community
- Price
- Free software
- Inference
- Your own provider key. Billed by your provider.
- Hosted workspaces
- Self-hosted
- Members
- Self-managed
- Concurrent hosted runs
- Self-managed
- Model requests in flight
- Your own capacity
- Hosted evidence retention
- Self-managed
- Hosted artifact quota
- None
- Hosted worker compute
- None
- Private run comparisons
- Local
- Policy canaries
- Local manual configuration
PRO(proposed)
- Price
- USD 29 / month
- Inference
- Your own provider key at launch. Billed by your provider, not included in the plan.
- Hosted workspaces
- 1
- Members
- 3
- Concurrent hosted runs
- 2
- Model requests in flight
- 8 per workspace
- Hosted evidence retention
- 30 days
- Hosted artifact quota
- 5 GB
- Hosted worker compute
- 10 worker-hours / month
- Private run comparisons
- Yes
- Policy canaries
- View reviewed releases
Max(proposed)
- Price
- USD 99 / month
- Inference
- Your own provider key at launch. Billed by your provider, not included in the plan.
- Hosted workspaces
- 5
- Members
- 10 in total
- Concurrent hosted runs
- 5
- Model requests in flight
- 32 per workspace, 64 per account
- Hosted evidence retention
- 90 days
- Hosted artifact quota
- 25 GB
- Hosted worker compute
- 50 worker-hours / month
- Private run comparisons
- Yes
- Policy canaries
- Managed canary workflow
Pricing and limits are subject to validation before paid launch. Prices exclude applicable taxes.
Open at the core.
The engine, CLI and benchmark generator are meant to be yours to read and run. Open-source release planned; no licence is published yet.
How billing is meant to work
Your provider, your inference bill
Every tier uses your own provider key. Your model provider charges you for tokens at its own prices. No plan includes model usage, and no plan promises unlimited compute.
What a hosted plan would cover
Workspaces, worker time, evidence storage and retention. One worker-hour is one provisioned worker running for one hour. It is not an agent-hour or a token amount.
Spending stays bounded
The engine reserves authorized spending before dispatch and stops at the ceiling you set. A spend ceiling is not a promise of a particular result.
Outside every plan
Experiments toward 10,000 participating agents are a separate research capacity program. Recursive self-improvement campaigns are separately budgeted research workloads.
The hosted product is in private development. See what exists today.