Deep Reasoning

AI funding and usage

AI included uses REDA credits. Bring your own AI uses your organisation’s provider account, billed directly by that provider. The licence selects one funding model for the organisation. In either model, REDA AI governs configured tools and actions in Salesforce. Set up your licence and AI accounts.

With AI included, select the agent’s Processing Tier. With Bring your own AI, select AI Connection and Model in the agent’s AI Account section, or inherit the organisation default. Different agents can use different supported providers and models. Prompts, tools, record context and channel permissions still define each agent’s job. Configure the agent’s AI account.

In customer-account mode, the selected model determines which Deep Reasoning levels and capabilities are available. Review the AI Account selector’s supported levels and warnings; do not assume every provider supports the same combination. Admin AI tools and document processing use the organisation default connection. Configure the default and test the workflow.

ⓘ

What is Deep Reasoning? Deep Reasoning lets you choose how thoroughly an agent works through a question before it replies. At the lowest setting the agent answers directly and quickly; at higher settings it takes more time to think a problem through step by step, which improves accuracy on complex, multi‑step questions — at the cost of a slower reply and more credits. It is set per agent and is Off by default.

Why it matters

Most everyday questions — “what’s the status of this opportunity?”, “summarise this case” — are answered well and instantly without any extra reasoning. But some tasks benefit from the agent slowing down and reasoning carefully: weighing several pieces of information, following a multi‑step chain, or making a judgement call. Deep Reasoning gives you a single dial to decide how much of that careful thinking an agent should do, so you can match each agent to the difficulty of the work it handles.

The trade‑off is simple: more reasoning means better answers on hard questions, but slower replies and higher credit usage. Because the right balance differs from one agent to the next, Deep Reasoning is configured on each agent individually rather than for the whole org.

The reasoning levels

The Deep Reasoning field is a picklist with six levels. Choose the lowest level that gives you the answer quality you need.

Level	Best for	Speed & cost
Off (default)	Everyday, high‑volume agents — lookups, summaries, straightforward Q&A, and most customer‑facing chat. The right choice for the large majority of agents.	Fastest, lowest cost.
Low	Questions that need a little extra care but should still feel quick — light comparisons or simple judgement calls.	Slightly slower, slightly more credits.
Medium	Genuinely multi‑step questions where accuracy matters — reasoning across several records, connecting related facts, or explaining a recommendation.	Noticeably slower, moderately more credits.
High	Complex analysis and decision support where a wrong answer is costly and a longer wait is acceptable.	Slower, higher credits.
Xhigh	The hardest, most involved tasks — deep, careful problem‑solving that chains many steps together.	Much slower, high credits.
Max	Maximum thoroughness, when getting the very best answer matters more than speed or cost. Use sparingly and test first.	Slowest, highest credits.

ⓘ

Start low, raise only if needed. Begin with Off. If an agent’s answers on genuinely complex questions aren’t deep enough, raise the level one step and re‑test. There is rarely a reason to jump straight to Max — the higher levels cost noticeably more credits and take longer, and a lower level is often just as good.

How to turn on Deep Reasoning

ⓘ

On an existing org, the Deep Reasoning field may not appear on your Agent Definition page until an administrator adds it to the page layout. If you don’t see the field, follow Adding the field to your page layout first.

Open the agent. Go to the Agent Definitions tab and open the agent you want to tune.
Find the Deep Reasoning field. It sits in the Context & Persona section of the record.
Choose a level. Pick Off, Low, Medium, High, Xhigh or Max from the picklist.
Save. The change takes effect on the agent’s next reply — there is nothing else to deploy.
Test. Ask the agent a representative hard question and check the answer quality and the response time. Adjust the level up or down until it feels right.

Where the setting applies

Deep Reasoning applies to everything a single agent does — its chat replies, the actions it takes, and its AI Recommendations on record pages. You set the depth once on the agent and it is used consistently across all of that agent’s work. Other agents are unaffected; each carries its own Deep Reasoning setting.

The Economy processing tier does not support Deep Reasoning. On an agent whose Processing Tier is Economy, every Deep Reasoning level behaves as Off — the setting is not rejected, it simply has no effect. If you need Deep Reasoning, keep the agent on Standard or Efficient. See Configuring Agent Settings.

Cost and performance

Higher reasoning levels use more credits per interaction and take longer to respond, because the agent is doing more work before it answers. Keep this in mind for high‑volume agents (for example a public website or community bot handling many conversations): leaving those at Off keeps them fast and economical. Reserve the higher levels for lower‑volume, higher‑value agents where the quality of a careful answer justifies the extra time and cost.

A good rule of thumb

Customer‑facing, always‑on agents → Off. Internal analysis or advisory agents that tackle complex questions for a smaller audience → Medium or High. Only go to Xhigh or Max for specialised, difficult work after testing that the added quality is worth the wait.

Adding the field to your page layout

When you update to a version that includes Deep Reasoning, the new field is added to your org automatically, but — as with any newly added field — it will not appear on your existing Agent Definition page until an administrator places it on the layout. Your agents keep working exactly as before in the meantime; the field simply behaves as Off until you choose otherwise.

Open Setup. Click the gear icon → Setup.
Go to Object Manager. In Setup, open Object Manager and find Agent Definition.
Open Page Layouts. Select Page Layouts, then open the layout your team uses (for example REDA AI Agent Definition Layout).
Drag the field onto the layout. In the field palette at the top, find Deep Reasoning and drag it into the Context & Persona section (next to Persona is a natural place).
Save the layout. Click Save. Administrators will now see and be able to edit the field on every Agent Definition record.

ⓘ

Who can edit it? Deep Reasoning is a configuration field for administrators. Users assigned the REDA AI administrator permission set can see and change it; standard users can view it but not edit it. If a colleague can’t see the field even after it’s on the layout, check that they hold the administrator permission set.

ⓘ

Existing agents. Agents created before this update have no value set for the field. A blank value is treated exactly the same as Off, so existing agents keep their current, fast behaviour. You can optionally set them to Off explicitly so the field always shows a value. Any agent you create from now on defaults to Off.

Advanced: platform‑wide safeguard

Platform Settings, in the REDA AI admin app, carries a Deep Reasoning Max Tokens Cap. It is a hard ceiling on the combined output-token budget — the answer plus the extended thinking — for any agent with Deep Reasoning on. Each reasoning level adds headroom on top of the base budget; this cap bounds the total so a reply still finishes inside the callout window.

The default is 24,000, and the setting has a floor of 4,000: enter anything lower and REDA AI uses 4,000. It is not optional in the sense of being off by default — the ceiling always applies; what is optional is changing it. The default suits virtually every org, so leave it alone unless REDA AI support advises otherwise. Raising it makes the longest replies longer and more expensive without making them more accurate.

Frequently Asked Questions

They control different things. Token Budget limits how long a single answer can be. Deep Reasoning controls how carefully the agent thinks before it answers. You can use both together — for example, a concise but well‑reasoned assistant.

No. Every agent stays at Off (or blank, which means the same thing) until you deliberately raise the level. Nothing about your current agents changes until you choose a higher setting on a specific agent.

Start with Off. It gives fast, accurate answers for the vast majority of questions. Only raise the level for an agent that regularly handles genuinely complex, multi‑step questions where you can see the answers would benefit from more careful thinking — and raise it one step at a time.

Yes. Higher levels do more work per interaction, so they use more credits and take longer to respond. That’s why Off is the default and the recommended choice for high‑volume agents. Reserve the higher levels for lower‑volume agents where the extra answer quality is worth it.

Yes. The reasoning level you set on an agent is used for that agent’s chat, its actions, and its AI Recommendations. You configure it once, per agent.

On an existing org, a newly added field doesn’t appear on your page automatically. An administrator needs to add it to the Agent Definition page layout — see Adding the field to your page layout. Also confirm you hold the REDA AI administrator permission set.

Next Steps

See Configuring Agent Settings for the full list of agent fields, or AI Recommendations to put an agent’s reasoning to work on your record pages.

← Previous: Configuring Agent Settings

Next: Build Agent with AI →