# Limits and quotas

Plan limits, resource caps and monthly usage allowances across Hobby, Pro, Team and Enterprise.

Every organization starts on the **Hobby** plan ($0 a month, plus usage). **Pro** ($15 a month, plus usage) raises solo limits and unlocks larger environments and custom model endpoints. **Team** ($30 a builder a month, plus usage; viewers and auditors are free) adds collaboration, organization policies, 400-day audit logs and limits that kosem raises on request. **Enterprise** starts from Team's limits with custom terms.

To see what an organization uses against its limits, open **Billing** › **Quotas** in the console or call [`ListQuotaInfosInOrganization`](/reference/api/quota-infos/list-quota-infos-in-organization).

## Your account [#your-account]

Each account can create **1 organization on Hobby** and up to **20 passkeys**. Paid organizations (Pro, Team and Enterprise) don't count toward the Hobby organization limit.

## Organization and access [#organization-and-access]

| Resource                                                   |             Hobby |               Pro |           Team |
| ---------------------------------------------------------- | ----------------: | ----------------: | -------------: |
| Projects                                                   |                 3 |                15 |  50 (raisable) |
| Members, guests and pending invitations                    |                 1 |                 1 | 500 (raisable) |
| Groups                                                     |                 0 |                 0 | 100 (raisable) |
| Service accounts                                           |                 5 |                25 | 200 (raisable) |
| Keys per service account                                   |                10 |                10 |             10 |
| Grants per invitation                                      |                20 |                20 |             20 |
| Organization policies                                      |     View defaults |     View defaults |       Editable |
| Audit logs (Admin Activity, Data Access and System Events) | Recorded (locked) | Recorded (locked) |       400 days |

On **Team**, billable builder seats count active members and guests who have any non-viewer role (directly or through a group). Read-only viewers, site viewers, billing admins, member admins and audit log viewers are free.

## Agents and sites [#agents-and-sites]

| Resource                                      | Hobby |    Pro |               Team |
| --------------------------------------------- | ----: | -----: | -----------------: |
| Concurrent agents                             |    10 |     32 |     100 (raisable) |
| Agent web page reads per minute (per project) |    10 |     30 |      60 (raisable) |
| Agent file storage (organization total)       | 1 GiB | 25 GiB | 250 GiB (raisable) |
| Sites                                         |     5 |     50 |     200 (raisable) |
| Versions kept per site (beside live version)  |     5 |     20 |                 50 |

- **Agent files:** up to 50 MiB per file, up to 1 GiB per agent, bounded by the organization's **File storage** quota (`files.storageGib`).
- **Sites:** up to 2,000 files, 25 MiB per file and 100 MiB total per site version.

## Data retention [#data-retention]

Plan limits define the longest kosem keeps each kind of data. On **Team** and **Enterprise**, Owners and Policy Admins can shorten retention for an organization or individual projects in **Settings** › **Data retention**.

| Resource                                          |        Hobby |           Pro |                           Team | Enterprise |
| ------------------------------------------------- | -----------: | ------------: | -----------------------------: | ---------: |
| Inactive agents, transcripts and attached files   |      90 days |      365 days | Unlimited (shortenable 7–365d) |     Custom |
| Inactive sites and published versions             |      30 days |     Unlimited | Unlimited (shortenable 7–365d) |     Custom |
| Stopped environments and disks                    | Up to 7 days | Up to 30 days |  Unlimited (shortenable 1–30d) |     Custom |
| LLM Gateway request log (daily usage totals stay) |       7 days |       30 days |    90 days (shortenable 1–90d) |   90+ days |
| Audit logs: Admin Activity and System Events      |     400 days |      400 days |                       400 days |   400 days |
| Audit logs: Data Access                           |      30 days |       30 days |               30 days (1–400d) |  400+ days |
| Audit logs: Policy Denied                         |      30 days |       30 days |                        30 days |    30 days |
| Soft-deleted projects (grace period before purge) |      30 days |       30 days |                        30 days |    30 days |
| Unpaid stopped environments (grace after $0)      |      14 days |       14 days |                        14 days |    14 days |

## Environments [#environments]

| Resource                   |                         Hobby |                             Pro |                            Team |
| -------------------------- | ----------------------------: | ------------------------------: | ------------------------------: |
| Concurrent vCPUs           |                             4 |                              16 |                   50 (raisable) |
| Concurrent memory          |                        16 GiB |                          64 GiB |              200 GiB (raisable) |
| Largest single environment | 2 vCPUs · 8 GiB · 50 GiB disk | 8 vCPUs · 52 GiB · 500 GiB disk | 8 vCPUs · 52 GiB · 500 GiB disk |
| Total environment disk     |                       100 GiB |                         500 GiB |            2,000 GiB (raisable) |
| Environments               |                            20 |                             100 |                  500 (raisable) |
| Prepared images            |                             1 |                              10 |                   25 (raisable) |
| Image storage              |                        10 GiB |                         100 GiB |              500 GiB (raisable) |
| Private registries         |                             2 |                              10 |                  100 (raisable) |

An agent's default environment is **2 vCPUs, 8 GiB memory and 10 GiB disk** (standalone environments default to **2 vCPUs, 8 GiB memory and 20 GiB disk**). An agent's environment stops 5 minutes after its turn and is deleted after 2 days stopped unless extended up to your plan's stopped retention cap (7 days on Hobby, 30 days on Pro, unlimited on Team and Enterprise). Stopped environments whose billing account has run out of credits are deleted after a 14-day grace period.

## LLM Gateway [#llm-gateway]

| Resource                             |        Hobby |      Pro |           Team |
| ------------------------------------ | -----------: | -------: | -------------: |
| Model providers                      |            2 |       10 |  25 (raisable) |
| Custom / self-hosted model endpoints | Not included | Included |       Included |
| Models                               |            5 |       25 | 100 (raisable) |
| Gateway request log retention        |       7 days |  30 days |        90 days |

| Limit                                                               | Value     |
| ------------------------------------------------------------------- | --------- |
| Projects a provider or a model is shared with, when not all of them | 500       |
| Gateway requests a minute, per project                              | 1,000     |
| Request body, files included                                        | 32 MiB    |
| Time until a streamed answer starts                                 | 5 minutes |
| Time a request lasts                                                | 1 hour    |

**Gateway requests a minute** counts the requests your own software sends to a project's models, through the provider endpoints or kosem's own format, on every plan. Agents' model calls don't count. A request past it gets `429` with `Retry-After`, the seconds until it fits ([errors](/reference/errors)).

Your model providers' own rate limits apply to every request routed through the LLM Gateway, agents' included. kosem doesn't limit tokens, and doesn't throttle model request speed by plan.

## Secrets [#secrets]

| Resource                                   | Hobby | Pro |             Team |
| ------------------------------------------ | ----: | --: | ---------------: |
| Secrets                                    |    25 | 100 | 1,000 (raisable) |
| Versions per secret                        |   100 | 100 |              100 |
| Connections (GitHub, AWS and Google Cloud) |     2 |  10 |    25 (raisable) |

## Monthly allowances and usage prices [#monthly-allowances-and-usage-prices]

Hobby, Pro and Team share the same usage list prices, paid from credits. **Model tokens** are billed directly by your providers with zero markup, and **agent model calls** are never charged per request.

| Usage                                 | Hobby free / month | Pro free / month | Team free / month | Past allowance        |
| ------------------------------------- | -----------------: | ---------------: | ----------------: | --------------------- |
| Agent model calls                     |          Unlimited |        Unlimited |         Unlimited | $0                    |
| Direct LLM Gateway API calls          |             50,000 |          250,000 |         1,000,000 | $0.10 per 1,000       |
| Agent web searches                    |                100 |              500 |             1,000 | $0.012 each           |
| Environment vCPUs (running)           |                  — |                — |                 — | $0.0504 per vCPU-hour |
| Environment memory (running)          |                  — |                — |                 — | $0.0162 per GiB-hour  |
| Environment disk (running or stopped) |                  — |                — |                 — | $0.10 per GiB-month   |

## Your usage limit [#your-usage-limit]

On Hobby, Pro and Team, owners and Billing Admins can set a **monthly usage limit** in **Billing** › **Overview** to stop paid usage (including cloud environments and direct gateway API calls past the allowance) when the month's charges reach the cap. You can also cap credit **auto-reload** per month and set per-project model spend limits in the LLM Gateway.

## Raising a limit [#raising-a-limit]

On **Hobby** or **Pro**, upgrading raises limits immediately. On **Team** or **Enterprise**, owners and Billing Admins can request higher limits from **Billing** › **Quotas** in the console or with [`CreateQuotaIncreaseRequest`](/reference/api/quota-increase-requests/create-quota-increase-request).
