> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reliantlabs.io/llms.txt
> Use this file to discover all available pages before exploring further.

# FAQ

> Answers to common questions

## Workflow Control

### How do I pause a running workflow?

Click the stop button or press `Esc`. The workflow stops with state preserved. Send a new message or click resume to continue.

### Can I change parameters while running?

Yes. Open parameters panel (gear icon), change values. Changes take effect on the next iteration. Common updates: switching to manual mode, adjusting temperature, changing model.

### What happens when a workflow fails?

Reliant pauses instead of crashing. Fix the issue (permissions, network, etc.) and click resume to retry.

## Threads and Branching

### How do I branch a conversation?

Click the branch icon on any message. A new chat is created with history up to that point.

### Can I switch workflows after branching?

Only before sending the first message. After that, the workflow is locked.

### How do I message a specific sub-agent thread?

Pause the workflow, select the target thread from the selector, then send your message.

### What's inherit vs new vs fork?

Thread modes control how sub-agents share conversation context. See [Threads](/features/threads) for a full explanation with examples.

## Workflows and Presets

### What's the difference between workflow and preset?

* **Workflow** = execution pattern (what steps, what order)
* **Preset** = parameter values (model, temperature, prompt)

### Why can't I change workflows mid-conversation?

Workflows define structure. Changing mid-run would break execution. Branch from earlier and select a new workflow before sending.

## Response Tools

### What are response tools?

Synthetic tools that force structured output from the LLM:

```yaml theme={null}
response_tool:
  name: review_result
  options:
    approved: "Code looks good"
    needs_changes: "Issues found"
```

Output is always `{ choice: string, value: string }`.

### When should I use them?

When you need to branch on structured data—code review pass/fail, classification, validation results.

## Data and Privacy

### Where is my data stored?

All local:

* **Conversations**: `~/Library/Application Support/reliant` (macOS) or `~/.config/reliant` (Linux)
* **API keys**: System keychain
* **No cloud sync** — anonymous usage telemetry is on by default (opt out in Settings > Privacy)

### Are conversations sent to Reliant?

No. Reliant Labs has no servers receiving your data. Only your AI provider sees messages for inference.

## Known Limitations

Current limitations we're aware of. We're working on these—check back for updates.

### Terminal

**Interactive utilities don't work.** Commands like `vim`, `nano`, and other interactive terminal programs aren't supported yet. Use GUI editors or non-interactive commands.

### Models

**Switching models mid-chat may cause issues.** Changing models during a conversation can break context, especially with thinking models. Workaround: branch first, then switch models on the new branch.

**Gemini may experience issues.** Google's Gemini models have intermittent compatibility issues. If you encounter problems, try switching to Claude or OpenAI models.

### Branching

**Branching from a different workspace uses the original workspace.** If you branch a message that originated from a chat in a different workspace, the new branch will be created in that original workspace, not your current one.

### Workflows

**Approval denial restarts the workflow.** Denying an approval node will cancel the entire workflow. When you resume, it starts over from the beginning. We're investigating better mechanics here.

**Tool access is enforced by permission level.** Tools are gated by permission levels (readonly, mutating, orchestrator) at execution time. If the LLM requests a tool that exceeds the granted permission, the call is denied with a clear error. This prevents unauthorized tool use regardless of what tools the LLM has seen in context.

**`continue_as_new` uses newer workflow versions.** When a workflow continues as new, it may pick up a newer version of the workflow definition, causing inconsistent behavior.

**Spawned agents limited to `builtin://agent` workflow.** The spawn node currently only works with the built-in `builtin://agent` workflow, not custom workflows.

### Experimental Features

**Activity viewer is in alpha.** The activity viewer is experimental and not fully supported. Expect rough edges.

***

**More help**: [Troubleshooting](/help/troubleshooting) • [Feedback](/help/feedback)