# AI platform execution limits

AI execution limits bound model and function work, while specific caches and compatibility fallbacks help selected operations continue.

*Release 3.3 · admin*

**Settings › AI › Accounts** · **Settings › AI › Features**

## What it is

AI execution limits bound how much work an AI feature can attempt. Different controls cover input size, repeated model calls, tool calls and elapsed time. Selected services also reuse cached information or retry a compatible request format.

## Who uses it and why

Administrators agree suitable limits with Powerfront so AI tasks have enough time and context to finish. Limits can also prevent repeated function work from continuing indefinitely.

## Where to find it

Powerfront manages shared limits in **Settings › AI › Accounts**. Features in **Settings › AI › Features** use those defaults and any applicable feature-specific context options.

You can access **Features** with the **Settings** role and **AI** access, or the **Developer** role.

## How it works

The default function iteration allowance is **3**, and **Max Function Seconds** defaults to **30 seconds**. An iteration can request several tools, so this count is not always the number of individual tool calls.

Managed context uses **Maximum run time (seconds)** instead of the older function-time default. Individual model requests and tools can have shorter deadlines. Specialist invocations can also have their own allowances.

Where an accumulated token allowance is supplied for a run, reaching it disables further tools after the generation is counted. Exceeding an individual tool-call allowance returns a tool-limit failure for the excess call.

Caches and fallbacks belong to particular services. For example, managed prompts can reuse cached prompt content, and selected providers can retry a compatible request format. These behaviours do not establish automatic switching to another model provider.

## Configure it

Ask Powerfront to review the shared account limits and the context mode your features use. Specify the task that stops too soon or runs too long so Powerfront can identify which limits to change.

The **Max Function Count** control offers 1 to 10, and **Max Function Seconds** offers 1 to 60 seconds. Managed **Maximum run time (seconds)** defaults to 1,800. **Maximum wake delay (seconds)** defaults to 300, and **Enable Continue Work** starts off.

Ask Powerfront to set **Maximum input tokens** when you need an explicit adaptive input budget. A blank value derives the budget from the model. See the context-management feature for compaction and recent-message controls.

### Settings

| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Max Function Count | Settings › AI › Accounts | website | 3 | Powerfront, on request | Sets the default iteration allowance for applicable AI execution loops. |
| Max Function Seconds | Settings › AI › Accounts | website | 30 seconds | Powerfront, on request | Sets the default elapsed-time allowance for applicable execution loops using simple context. |
| Maximum input tokens | Settings › AI › Accounts › Advanced context settings | website | Not set: derived from the model limit | Powerfront, on request | Overrides the adaptive context input budget; a blank value derives it from the model limit. |
| Maximum run time (seconds) | Settings › AI › Accounts › Advanced context settings | website | 1800 seconds | Powerfront, on request | Sets the default managed run duration, subject to shorter provider and tool deadlines. |
| Maximum wake delay (seconds) | Settings › AI › Accounts › Advanced context settings | website | 300 seconds | Powerfront, on request | Bounds requested delays between continued-work turns. |
| Enable Continue Work | Settings › AI › Accounts › Advanced context settings | website | Off | Powerfront, on request | Allows managed agent runs to continue working and send interim updates when enabled for the feature. |

### Permissions

| Permission | Lets you |
|---|---|
| AI | With the Settings role, allows access to feature-specific AI options; the Developer role can also access Features. |

## Limits and good to know

- Input context, generated output and accumulated run tokens are different quantities. One setting does not cap all three.
- Limits bound execution; they do not create a monetary spending cap or a guaranteed response time.
- A timeout or cancellation does not establish that an external action already performed by a tool has been undone.
- Cache freshness and fallback behaviour vary by service. There is no single cache duration for all AI work.
- A provider compatibility retry does not guarantee recovery from an outage.

## Depends on and works with

### This feature uses

- **Configured in** [AI feature settings and change history](AI-001.md): Max Tokens on the AI feature sets a feature-specific allowance or inherits the account default when zero.
- **Configured in** [AI run limits](AI-014.md): The feature's Max Tokens and its account's function allowances control standard text-agent runs.
- **Configured in** [Context Compiler for AI conversations](AI-015.md): Managed context and continued-work settings determine applicable run and wake-delay limits.
- **Configured in** [Specialist sub-agents for AI tasks](AI-023.md): Specialist invocations can impose their own smaller execution allowances.
- **Configured in** [AI accounts and shared defaults](AIP-001.md): Powerfront manages account-wide defaults shared by features using that account.
- **Configured in** [AI model catalogue](AIP-002.md): Model context limits affect the available token budget.
- **Configured in** [External AI prompts and versions](AIP-014.md): External prompt selections set their own cache duration and fallback instructions.

## FAQ and troubleshooting

### Why did the AI stop before finishing all its tools?

It may have reached an iteration, tool, token or time allowance. Give Powerfront the feature and conversation so they can identify the effective limit.

### Why does a managed run last longer than Max Function Seconds?

Managed context uses its run-time allowance. The older function-time default does not describe every execution mode.

### Will INSIDE switch providers if one fails?

The specific compatibility fallbacks described here do not provide general cross-provider failover. Ask Powerfront to confirm the recovery behaviour of your selected feature and provider.

### Why did a shared limit change several features?

Those features may use the same AI account. Ask Powerfront to review the affected account and any feature-specific overrides.