FAQ
Frequently asked questions. For detailed information, see the referenced chapter.
Agents
Yes — HuLoop creates a new version each time changes are saved, allowing any previous version to be reviewed, compared, or restored. See Templates and Versions.
Pin a version as Live in the agent’s Versions table. Manual runs will use the pinned version until a different version is selected. See Templates and Versions.
Use Export as JSON on BUILD → Agents, then + New Agent → Import JSON in the target space (recreate any tools and knowledge bases first).
Yes — enable the invoke_agent tool, or chain agents. See Tools and Collaboration.
A single run processes one request and returns a result. A conversation retains previous interactions, allowing the agent to reference earlier messages.
Running & Results
The run stops with a budget_exceeded status and keeps what it produced so far; raise the Max Budget and run again. See Running and Results.
Verify that the required tools are enabled (blue chips), the prompt includes instructions for the intended task, and no guardrails are preventing tool execution. See When Something Goes Wrong.
Click Stop to cancel. If the status is waiting, the agent needs your input or an approval — open the run and respond. See When Something Goes Wrong.
Yes — click Resume on the failed run while its checkpoint still exists (about 7 days). Completed steps are skipped. See Task Flows.
Your current role does not permit this action. Contact an administrator if additional permissions are required.
Knowledge Bases
.txt, .md, and .pdf, up to 10 MB each. See Knowledge Bases.
Automatically — it finds the most relevant information and adds them to its context before answering. See Knowledge Bases.
Yes — upload new versions or delete individual documents at any time.
Tools & Safety
These tools run in a sandboxed environment with restricted file access and blocked high-risk commands. Sensitive credentials are removed automatically, and tool access can be restricted further through guardrails. See Safety and Guardrails.
No — requests to private, localhost, and cloud-metadata addresses are blocked automatically. See Safety and Guardrails.
Yes — pick only the tools it needs and add an “Allowed/Blocked Tools” guardrail that’s enforced even if a tool is technically available. See Safety and Guardrails.
All data is isolated to your organization’s tenant and is not accessible to other organizations. See Safety and Guardrails.
An administrator determines which AI models are available and configures the default model. In most cases, the default model can be used. For information about model availability, provider credentials, and model restrictions, see the Admin Reference — Administration.
