Docs

Sovariq modes and thinking levels

Ask, Plan, Build and Ideate decide what Sovariq may do; Low to Ultra decide how hard it thinks. How each works, how Sovariq switches between them, and what each costs.

Updated Oct 20, 2026

Four modes

ModeReads your projectChanges itEnds with
AskYesNeverAn answer, with links to the pages it used
PlanYesNeverA proposal (tables, who can access what, functions) and Build this plan
BuildYesYes, after your OKThe change, verified, with an undo in Version History
IdeateYesNeverIdeas, priorities and risks, and Plan the first step

Ask

For questions: "who can read the orders table?", "why did this request fail?", "how do webhooks retry?". Sovariq reads your project, the docs and its playbooks, and answers. It never changes anything in Ask.

Plan

SovariqPlanThinking: High

Plan a referral program: each user gets a code and a reward when a friend signs up.

schema_digestRead users, orders and the auth hooksDone

Plan: a referral_codes table (one per user, owner-only), a referrals table written by a sign-up hook, and a credit_referral() function. Nothing is changed yet.

Press Build this plan when it looks right.

Build

Does the work, step by step, and checks each step. Writes wait for your approval unless you turned on Auto-run safe changes (the menu under the message box, or Sovariq › Settings); deleting data, running code and anything that leaves the project always ask (Sovariq safety, approvals and guard rails).

Ideate

Helps shape the product and the business: who it is for, the features worth building first, how it makes money, the risks. It works from your actual project, changes nothing, and Plan the first step turns an idea into a plan.

Switching modes by what you say

plan itbuild itnot yetIdeate"brainstorm features"Plan"plan this first"Build"looks good, go ahead"

The mode menu follows: "plan this first" or "don't build anything yet" in Build switches to Plan; "looks good, go ahead" after a plan switches to Build; "brainstorm features" or "how could we make money" switches to Ideate. Sovariq only enters Build from what you typed, never on its own.

How hard it thinks

The Thinking menu next to the message box (and Sovariq › Settings › General) sets the effort for each message. Higher levels think longer and get more room, so they are more accurate on hard work and use more tokens. Every level counts its real token use against your daily Sovariq allowance, or bills your own provider account when you use your own model.

LevelWhat it doesUse it forPlans
LowAnswers straight away, no extended thinking, short repliesQuick questions, lookups, small editsAll
Medium (default)Thinks to plan and lightly while building; answers questions directlyEveryday buildingAll
HighThinks on every step, hardest when planningTricky questions, debugging, policiesAll
MaxThinks hard everywhere, twice the steps, tool calls, time and reply length, and checks its work before it finishesA whole new backend, a large migrationPro and up
UltraThe deepest thinking, three times the room, and verifies each change with its tools (reads it back, dry-runs, simulates) before it finishesWork that must be right the first timeUltra, Team, Enterprise

If your plan doesn't include a level, the message runs at the highest level it does include and Sovariq tells you which. The per-run spending cap still applies at every level.

FreeProUltraTeamEnterprise
Sovariq Max thinking—✓✓✓✓
Sovariq Ultra thinking——✓✓✓

Answers that cost nothing

Common requests are answered without a model call, so they are instant and cost no tokens:

KindExamples
Where things are"where do I rotate the anon key?", then "open it"
Showing a table"show me the orders table"
Clear one-line schema edits"add a phone number to customers", "make the product SKU unique", "remove the legacy_code column from customers", "comments.post_id should point to posts", "add an index on orders.created_at"

An edit still runs as an ordinary tool call, with the same checks, approvals and audit trail. Anything less clear goes to the model.