Does Turning On Tool Discovery Lower My Agent's Credit Usage?

Last updated: June 5, 2026

Usually yes for tool-heavy agents, but not always. Tool Discovery now turns on automatically for custom agents when connected tool schemas would take up too much context, and you can still force-enable it from the agent's Abilities settings. For small agents, or agents that use most of their tools every run, the credit impact can be neutral or slightly higher, so compare actual run credits before deciding whether to force it on.

How Tool Discovery Changes Credit Cost

Every agent run includes the agent's tools in the context the model reads. Tool Discovery changes when those tool definitions are loaded, and that shifts where credits get spent:

With Tool Discovery off: the full definition of every tool you added is loaded into context at the start of each run. That's a fixed cost paid up front on every run — larger when the agent carries many tools — but the agent never spends extra calls fetching a tool's details later.
With Tool Discovery on: the agent starts with a small context and pulls in each tool's definition on demand, only when it decides it needs that tool. The upfront cost drops, but each discovery step the agent takes during the run uses some credits of its own.

So enabling Tool Discovery swaps a fixed upfront cost for a variable per-run cost. Whether that's a net saving depends entirely on the agent.

When Tool Discovery Turns On Automatically

For custom agents, Tool Discovery now uses auto behavior by default. When a conversation starts, Gumloop estimates how much of the model context would be consumed by connected tool schemas. If those schemas exceed roughly 10% of the available context, the agent switches to Tool Discovery automatically.

You only need to use the Tool Discovery toggle in the agent's Abilities section if you want to force Tool Discovery on even when the agent has a smaller tool set. This can be useful for very broad agents, but it is no longer required just to avoid oversized tool context.

When It Saves Credits vs. When It Can Increase Them

It typically lowers credits when:

The agent carries many tools (a large, multi-integration tool set), so the upfront context it avoids is significant.
Each run only uses a small subset of those tools — the agent skips loading everything it won't touch.

Tool-heavy agents in this situation commonly see meaningful per-run reductions.

It can slightly increase credits when:

The agent is already lean — it has only a handful of tools, so there wasn't much upfront context to save, and the on-demand discovery steps cost more than they save.
The agent uses most of its tools on every run, so it ends up discovering nearly all of them anyway — paying the per-run discovery cost on top of work it would have done regardless.

On agents like these, the change can be a wash or a small net increase. This is expected behavior, not a bug — it's the natural tradeoff of loading tools on demand.

How to Tell If It's Actually Helping Your Agent

Don't assume either direction — measure it on the specific agent:

Pick a representative task the agent runs often, and run it a few times with Tool Discovery off. Note the credits each run used from your Run History.
If Tool Discovery has not already activated automatically, turn it on in the Abilities section of the agent's configuration and save. Custom agents automatically switch to Tool Discovery when connected tool schemas exceed roughly 10% of the model context window, so the toggle is mainly for force-enabling it on smaller tool sets.
Run the same task a few more times and compare the per-run credits against your earlier numbers.
Keep Tool Discovery on only if the per-run credits actually dropped. If they went up or stayed flat, turn it back off for that agent.

Run a few times each way rather than once — credit cost per run varies with the task, so a single comparison can mislead.

If you ask the agent itself to estimate the impact, treat that as a directional hint only. The real proof is the before/after credits on actual runs, not the estimate.

What Tool Discovery Actually Does

Beyond cost, Tool Discovery also changes how your custom agent loads tools: instead of putting every connected tool schema into the model context upfront, the agent lazy-loads tool schemas only when they are relevant. Custom agents now use this automatically when the connected tool schemas exceed roughly 10% of the model context window, and the Abilities toggle force-enables the same behavior even when the tool count is smaller. The agent still respects per-tool restrictions and disabled servers you've configured. Full details are in the Agents documentation. Note that /chat (general) agents always use Tool Discovery.

How Do I Know It Worked?

After enabling Tool Discovery and re-running your representative task, the per-run credit figures in Run History should be lower than your baseline. If they are, you're saving credits; if they aren't, the agent is one where Tool Discovery doesn't pay off and you can safely leave it off.

Still Need Help?

If this didn't resolve your issue, reach out to support at support@gumloop.com.