Limits
This page lists the caps you can run into while using Workprentice: how many agents a workspace holds, how long an MCP call waits, how large a webhook body may be, and the rest. Each row links to the page that explains the limit in context.
Some caps are policies, which can be set per organization, workspace, or agent. The Default column is the value that applies until an admin sets one. The Pro and Enterprise columns are how far an admin can raise the limit on that plan: whatever is set, or the Default, is held to that ceiling. A ceiling below the Default therefore lowers it, as with Pro’s 10 workspaces and its 25 fresh A2A turns an hour. A dash means the plan does not change the limit. The last section says where each policy is set.
Workspace
Section titled “Workspace”| Limit | Default | Applies to | Pro | Enterprise | Details |
|---|---|---|---|---|---|
| Workspaces | No cap | Organization | 10 | No cap | Organizations and Workspaces |
| Agents | No cap | Workspace | 100 | No cap | Guardrails |
| Assignments | No cap | Workspace | 500 | No cap | Guardrails |
| Monthly spend | No cap | Organization or workspace | — | — | Guardrails |
| Skills per import | 25 at a time, from a source with at most 50 skills | Import | — | — | Skills |
Sessions
Section titled “Sessions”| Limit | Default | Applies to | Pro | Enterprise | Details |
|---|---|---|---|---|---|
| Session budget | None | Session, including its subagents | — | — | Guardrails |
| Repeated identical tool calls | Warned after 3, turn ended after 5 | Session | — | — | How Agents Run |
| Conversation compaction | Older history is summarized as a long conversation grows; the model’s full context window is available on every plan | Session | — | — | How Agents Run |
| Shell command time | 2 minutes unless the agent asks for longer, up to 60 minutes | Command | — | — | How Agents Run |
| Files attached to a message | 5, each up to 4 GiB; files over 10 MB (images over 3.5 MB) stay in storage until the agent fetches them | Message | — | — | How Agents Run |
| Work product file | 4 GiB | Work product | — | — | Work Products |
| Pasted text | 10 blocks of up to 100,000 characters | Message | — | — | Sessions |
Agents
Section titled “Agents”| Limit | Default | Applies to | Pro | Enterprise | Details |
|---|---|---|---|---|---|
| Home directory size | 20 GiB disk, caches included | Agent | — | — | Memory and Storage |
| Delegation depth | 1 (a subagent cannot start its own) | Tree of delegated work | 2 | 3 | How Agents Run |
| Subagent result | 50,000 characters | Subagent, returned to its parent | — | — | How Agents Run |
| Subagents running at once | 5 | Parent session | 5 | 10 | How Agents Run |
| Consultations | 10 | Turn | 10 | 25 | How Agents Run |
| Agent API relay hops | 2 | Chain of agent API calls | 2 | 3 | Give an Agent an API |
Channels
Section titled “Channels”| Limit | Default | Applies to | Pro | Enterprise | Details |
|---|---|---|---|---|---|
| Agent posts | 4 per 10 minutes | Agent, per channel | — | — | Channels |
| Message length | 16,384 characters | Post | — | — | Channels |
MCP, agent API, and A2A
Section titled “MCP, agent API, and A2A”| Limit | Default | Applies to | Pro | Enterprise | Details |
|---|---|---|---|---|---|
| Fresh agent turns over MCP and agent APIs | 100 per sliding hour | Workspace | — | — | Connect from MCP Clients |
| MCP tool result | 50,000 characters | Call | — | — | Connect from MCP Clients |
| MCP wait for a reply | 60 seconds for ask_agent, up to 300 |
Call | — | — | Connect from MCP Clients |
| Agent API call | Waits 5 minutes, returns up to 50,000 characters | Call | — | — | Give an Agent an API |
| Agent API access token | 1 hour | Token | — | — | Give an Agent an API |
| Agent API refresh token | 30 days | A person’s connection | — | — | Give an Agent an API |
| A2A sessions at once | 10 | Workspace | 10 | 50 | Delegate over A2A |
| Fresh A2A turns | 50 per sliding hour | Workspace | 25 | 100 | Delegate over A2A |
| A2A streams at once | 20 | API key, counted on each server separately | — | — | Delegate over A2A |
Webhooks and hooks
Section titled “Webhooks and hooks”| Limit | Default | Applies to | Pro | Enterprise | Details |
|---|---|---|---|---|---|
| Webhook body | 5 MiB | Request | — | — | API Keys and Webhooks |
| Hook write-backs | Once per effect, per ticket | Ticket | — | — | Hooks |
| Chained assignments | 2 hops past the original trigger | Chain | — | — | Hooks |
Changing a policy limit
Section titled “Changing a policy limit”The workspace, agent, and assignment counts, monthly spend, session budget, and loop thresholds are set on the Limits & budgets tab of organization and workspace settings; an agent’s Guardrails card carries the session budget and loop thresholds for that agent alone.
The fan-out limits (delegation depth, subagents at once, consultations per turn, and agent API relay hops) are set under Collaboration, and the A2A caps under Limits & budgets, in organization and workspace settings. Each field shows its default and how high your plan lets you set it, and a value above the plan’s ceiling is refused when you save.
The remaining policy caps have no settings screen. Raising one means writing
its policy through the app’s API with
PUT /api/v1/{scope}/{id}/policies/{policy}, where the scope is orgs (for
an organization owner or admin), workspaces (for a workspace admin), or
agents (for anyone who can edit the agent). The nearest scope that sets a
field wins, and the plan’s ceiling applies to API writes too.
agent_collaboration:channel_post_rate, and the fan-out limits above for a single agent.external_protocols:mcp_cold_starts_hourly, the cap shared by MCP and agent APIs.
A write replaces that scope’s whole policy, so send every field you want to keep.
For the complete site index, see llms.txt.