Mode and depth

Two controls

The mode says what kind of work a turn is: build, chat, research, a long build, or a team. The depth says how hard Mori thinks while she does it. They are set apart, so a research turn can think deeply and a build can stay quick.

Depth: Instant to Max

Depth is a slider with five notches, titled Depth, fast to deep. It sits in the model chip's menu (the chip reads, for example, "Opus 5.5 · Balanced"), and /depth sets it.

• Instant: the quick model, no thinking. Good for small asks. Haiku 4.5.
• Quick: Sonnet 5.5. Light thinking on harder asks, little or none on ordinary ones.
• Balanced, the default: Opus 5.5. Thinks as much as the ask needs.
• Deep: thinks longer on every step. Uses more of your plan.
• Max: Opus 5.5 at full effort. Slowest, for the hardest problems.

The models above are what Auto picks on your own OpenRouter or Anthropic key. Pick a model by hand and it stays pinned: the slider then moves only how hard that model thinks and how many steps it may take.

Depth is a ceiling

Each notch sets the most a turn may think, not the least. A greeting at Max still answers quickly, and an ordinary ask runs a level under the ceiling. Plain chat uses the quick model on every notch; from Balanced up, a question that needs reasoning moves to the notch's model.

Say "think hard", "think harder" or "ultrathink" and that one turn goes up a notch. When two checks fail in a row at the top of the current notch, the thread moves up one.

Depth on a plan

On a plan's credits, every paid plan runs the same models at each notch. Free on Mori's own models runs fast models, starts at Quick, and has Deep and Max locked: they come with every paid plan.

Paid plans have a daily count of Deep and Max turns: 30 on Plus, 100 on Pro, 300 on Studio, and no limit on Max. When the day's turns are used, a turn at Deep or Max runs at Balanced until midnight, and Mori says so. See Plans, credits and your own keys.

Modes

Build is the mode you work in: Mori researches, plans, builds and checks in one go. The other modes are a choice for when you want less, or more:

• Spark: chat only. Quick replies, changes nothing.
• Trance: research only. Reads and cites, never writes.
• Forge: long builds, with every tool loaded at once.
• Swarm: a team of agents builds parts at the same time, up to eight.

Pick one from the mode menu, or with /stage. A mode word and a colon at the start of a message switches the mode for that one message:

spark: what does this error mean?trance: compare the three sync libraries and cite your sourcesforge: port the sync engine to the new schema and prove the tests pass

Spend presets

Beside the depth, a spend preset decides which model does the smaller jobs around the work:

• Saver: the cheapest model wherever it is safe.
• Balanced, the default: the quick model for chat, your model for the work, and a local or cheap one for housekeeping.
• Full: your model everywhere.

The jobs it covers are building, research, chat, housekeeping, helpers and double-checking.