The model

The model that runs a business,
not a conversation.

nallo-1 is the planning and execution model behind Nallo. It is not tuned to answer well — it is tuned to take an ordered chain of reversible actions that never breaks a rule you set.

nallo-1 · in production since Q2 2026

nallo-1, the model behind Nallo
0.4 sintent → action
~250decisions / business / day
200markets scored
100%actions reversible

How it is built

Five stages, one loop.
The verifier feeds the planner.

PerceptionReads your real tools through OAuth — store, ads, payments, support.
World modelKeeps a live state of the business: margin, stock, spend, cash, sentiment.
PlannerSearches action chains under your hard constraints. Rejects before it proposes.
ExecutorCommits reversible steps, one at a time, with a compensation path for each.
VerifierChecks the result against the prediction. Feeds the error back into the planner.

What makes it different

Most agent models optimise for
task success. This one optimises
for not breaking your rules.

01

Constraint-first, not reward-first

A plan that earns more but violates your margin floor is a failed plan, not a trade-off. Constraints are checked during search, not after — so an unsafe action is never proposed in the first place.

02

Reversibility is a design constraint

nallo-1 will not commit a step it cannot undo or compensate. That rules out some fast paths, and it is why a bad night costs you an apology instead of a business.

03

Verification is continuous

Every action carries a prediction. The verifier compares it to reality and the gap becomes training signal for the next plan — nightly, on your own data, never pooled with anyone else’s.

Specifications

Measured on production workloads,
not on a benchmark suite.

Intent → action, median0.4 s
Planning horizon30 days, replanned nightly
Constraintshard: margin floor, budget ceiling, cooldown, jurisdiction
Reversibilityevery committed action reversible or compensable
Business state in contextcatalogue, orders, ad accounts, settlements, support threads
Markets scored200, continuously
Content languagesEN / FR / ES · DE, IT in pipeline
Decisions per business, per day~250
Verificationnightly health check + Dreaming simulation
DeploymentEuropean infrastructure, GDPR by design

Three variants

One family, three budgets.

nallo-1

The production model. Runs the chain end to end, proposes and commits actions under your rules.

Default on every plan

nallo-1-mini

Distilled for the high-frequency loops — bid adjustments, support triage, stock alerts. Sub-100 ms, no planning.

Always on, no extra cost

nallo-1-dream

The offline sibling. Runs during Nallo Dreaming with a far larger search budget, because nothing is live.

Nightly, 01:00–05:00

What it does not do well yet

We would rather write this
than let you discover it.

01

Attribution

Reallocating spend is only as good as the attribution underneath it. Post-cookie, this is our weakest input and we treat its output as a prior, not a fact.

02

Long-horizon credit

Knowing which decision from three weeks ago produced today’s margin is largely unsolved. Dreaming helps; it does not close the gap.

03

Anything physical

nallo-1 operates what it can reach through an integration. Logistics, manufacturing and anything requiring a human signature stay yours.

More detail on the research page.

Ten founding merchants.
Free for life.

GetNallo is not open to everyone yet. Applications are reviewed within five working days.

Closed beta — 7 of 10 spots left

Free for life for the ten founding merchants.

Apply for the beta