VDF AI Networks FAQ

Running networks

My network is taking a long time. Is it stuck?

Probably not. A typical network runs in a few minutes, but research-heavy or multi-source ones can take longer. Two checks before you assume something’s wrong:

• Is there an active stage indicator? If yes, it’s still working.
• Has any intermediate output appeared? If yes, the run is making progress.

If you’ve waited a few minutes past expected and no stage has updated, you can stop the run and start fresh — but more often than not, a brief pause solves it.

Can I edit the inputs mid-run?

You can pause a run, adjust the inputs or a specific stage, and continue from that point. You don’t have to start the whole network over. This is one of the most underused features — use it whenever an early stage drifts.

What happens if a stage fails?

The platform notifies you and stops the run. Most failures fall into three categories:

• Missing input — the network expected a source you didn’t provide.
• Source unavailable — a connected app is disconnected or a file was moved.
• Stage couldn’t make progress — usually because the inputs didn’t contain what the stage needed.

In each case, fix the underlying cause (provide the input, reconnect the app, give better source material) and rerun the stage.

Can I rerun just one stage?

Yes. You don’t have to rerun the whole network. Adjust the stage that needs to change, rerun from there, and the network continues with the updated output flowing into the next stages.

This is the most efficient way to refine: change the smallest thing, see the smallest effect.

Adjusting workflows

A stage isn’t producing what I want. What do I change?

Usually one of three things:

• The stage’s input is wrong. It pulled from the wrong source or got incomplete context. Check the inputs.
• The stage’s instructions are too vague. Tighten the description: name the audience, the format, the exclusions.
• The wrong specialist is running the stage. Some platforms let you change the specialist a stage uses; if not, you can split the stage into two with clearer instructions for each.

Can I add a stage to an existing network?

Yes. Open the network, add a stage at the position you want, and describe what it does. The platform will reconnect the surrounding stages so the data flow continues to make sense.

Can I remove a stage?

Yes. Make sure the stage before and after can still pass useful information without it — sometimes a removed stage breaks a dependency. Test the network on a real task after removing.

My network feels too long. What’s the fix?

A useful self-check on every stage: “If I removed this, would the final output get noticeably worse?” If no, remove it. Most great networks have three to six stages, not ten.

If you can’t trim any single stage but the whole thing feels heavy, consider splitting it into two networks chained together — see Building workflows.

Sharing and team usage

How do I share a network with my team?

Once a network is producing reliable results, save it to your team library. Give it a clear name (the deliverable, not the project), a one-paragraph description, and ideally an example run. Then tell someone about it — discovery matters as much as the save.

Can two people run the same network at the same time?

Yes. Runs are per-user and per-input set. Two people running the same network on different tasks get their own independent runs and outputs.

Can I edit a network someone else built?

Depends on how it’s shared. Most workspaces let you fork a network — make a copy you can edit without changing the original. The original owner can also grant edit access to specific people.

How do I see who’s using a network?

Most workspaces surface usage stats on shared networks — recent runs, total runs, common feedback. If yours doesn’t, check with your workspace admin.

Choosing the right tool

When should I use Chat instead of a network?

Use VDF AI Chat when:

• You’re exploring a problem and don’t know the shape of the work yet.
• The task is one-off — you won’t run this again.
• You need a fast, conversational answer rather than a polished deliverable.

When should I use an Agent instead of a network?

Use VDF AI Agents when:

• The task is a single deliverable in a known format.
• One specialist can produce the result.
• You don’t need stages, intermediate review, or multi-source reconciliation.

When are networks definitely the right call?

Networks are the right call when:

• The task has obvious stages (research → draft → critique → final).
• Multiple sources need to be combined into one output.
• The work repeats and the team benefits from a shared, consistent shape.
• The deliverable has a known format you reuse.

A test: if you’ve chained two or three agents by hand for the same recurring task on three separate occasions, that’s a network waiting to be built.

Sources and grounding

Can a network pull from different sources at different stages?

Yes. The whole network can have a primary source set, and individual stages can pull from additional sources scoped to them. For example: the network reads from a customer’s contract folder; the research stage also pulls from a public-data source; the critique stage references your team’s style guide.

What if a source has changed since I last ran the network?

The next run will pick up the current state of the source. Two things to watch:

• Connected apps may have re-authenticated. Confirm the connection is live.
• Source structure may have changed. If folders were renamed or moved, the network may need to be updated.

See Connecting sources for managing connected apps.

How do I make sure facts are grounded in my sources?

Two things help:

• Be explicit at the network level. “Only reference facts that appear in the attached sources. If something isn’t there, flag it.”
• Add a verification stage. A late-stage step whose only job is to check each named fact against the sources is a powerful safety net.

Triggers, routing, and advanced patterns

How do schedules work?

A network can be set to run on a calendar — every hour, every morning, weekly, monthly, or a custom cadence. You also pick where the output goes (email, Slack, a folder), so the result lands where someone will see it. Schedules are great for “this would be useful at 7 a.m. every weekday.” See Triggers and schedules.

What’s a sub-network?

A sub-network is a whole network used as a single step inside another network. Useful when you’ve already built a small, well-tested workflow for a narrow job (like “produce a tear-sheet from a customer file”) and want to drop it into a bigger workflow without rebuilding it. See Node types.

Can VDF AI pick the right model for me without my having to learn model names?

Yes — that’s what Smart routing does. The platform looks at each step, picks the right model from the catalog, and tells you what it chose. You can override per step when you have a strong opinion. See Smart model routing.

What’s the energy / sustainable mode for?

Sustainable mode tells the router to prefer models with lower energy and carbon impact among the ones that would do the job well. Most useful for high-volume scheduled networks and for organizations with sustainability targets. Quality stays high; cost usually drops too.

How do I keep a runaway network from blowing through our budget?

Set a budget cap on the network — per run, per day, or per month. The network stops cleanly at the cap and tells you what happened. See Policies and budgets.

Can I restrict which models a network is allowed to use?

Yes — set an allowed-models list on the network. The smart router will only pick from your list. Useful for compliance, for organizations with internal model policies, and for keeping a network’s behavior stable over time.

Still stuck?

• How networks work — the mental model in depth.
• Building workflows — turning a recurring task into a saved network.
• Triggers and schedules — match the right trigger to your workflow.
• Policies and budgets — set guardrails before sharing.
• Governance and admin — for workspace admins.
• Talk to us — if your question isn’t here, send it our way.