Free (Personal)
Retention + training requiredIndividual builders receive the workflow builder with standard rate limits in exchange for required ATOMS retention and internal-model training use under disclosed Free-plan terms.
Enterprise AI control plane
Own the intelligence layer before you own the model.
ATOMS turns fragmented model spend into governed workflow intelligence—then compounds the evidence, economics, and consented outcomes required to launch an owned enterprise middleware model within 24 months.
Policy firstRoute only to eligible endpoints
Task measuredQuality gates beat brand rankings
Provider neutralSwap models without rebuilding workflows
Evidence ownedOutcomes become the durable asset
The wedge
Middleware is the shortest path to model ownership.
A frontier-scale pretraining run is not a low-lift startup plan. A policy-aware proxy is. It can earn distribution, recurring gross profit, evaluation data, and enterprise trust before ATOMS funds a single owned-model workload.
The credible promise is frontier-competitive on selected middleware tasks—not best at everything on day one.
REFERENCE FOUNDATION · CATEGORY 1
Account profiles and privacy filters are separate controls.
The account determines commercial limits and margin. The privacy filter determines where data may go. Standard, Private, and ZDR are the primary routing filters—and carry no added ATOMS platform surcharge in the source concept.
Starter / Micro
3.0% reference marginCovers payment-processing overhead on small credit top-ups while preserving access to every privacy filter.
Standard Team
2.0% reference marginA shared wallet for unlimited members with a 25-session concurrency pool. Growth is welcomed; overlapping demand becomes the capacity signal.
Enterprise Organization
1.0% reference marginOrganization-level funding, nested team and user budgets, priority routing, and contracted capacity for dense agentic loops.
PRIMARY FILTERCost-first
Standard
No ATOMS retention restriction. Route across all commercially eligible public and private APIs for maximum token-cost savings.
- Public spot-market routing
- Standard provider pass-through rates
- Provider terms remain visible
PRIMARY FILTER$0 extra fee
Private
Provider retention for logging or debugging may be allowed, but training on customer data is strictly disallowed.
- Filters out training-eligible APIs
- Standard provider token rates
- Enterprise compliance focus
PRIMARY FILTER$0 extra fee
ZDR
Zero-day prompt retention. Route only to verified enterprise endpoints that do not log prompt content.
- Zero prompt logging requirement
- Standard provider token rates
- Strict-security focus
ACCOUNT ARCHITECTURE
One wallet. Nested budgets. No seat-count penalty.
Standard removes member limits so adoption can spread naturally. ATOMS monetizes verified AI activity and capacity: a 25-session shared pool makes simultaneous workflow demand—not adding a colleague—the transparent Enterprise upgrade trigger.
User
User → workflowsA personal balance and workflow allowance for one paid builder.
Team → User
Unlimited members · 25 concurrent sessionsAdd people without buying seats. Allocate the team wallet to user envelopes and govern how quickly each person can spend.
Organization → Team → User
Nested budgets · contracted capacityFund the organization, delegate team pools, cap users, and preserve consolidated controls and reporting.
PROVISIONING EXAMPLE$100
→Team wallet
10 PEOPLE$10
→Equal user envelope, or a custom cap
3 WORKFLOWS EACH30
→Potential simultaneous sessions
STANDARD POOL25 + 5 queued
Enterprise capacity becomes relevant only when the runs overlap.
CROSS-VERTICAL FLYWHEEL
Work proves the value. Life expands the habit.
Otter clients can discover ATOMS at work, apply the same approachable workflow patterns to personal tasks, and bring successful automations back into their teams. More governed usage improves templates, evaluations, routing evidence, and savings visibility.
- 01Workplace adoption
- 02Personal automation
- 03More verified outcomes
- 04Greater spend + capacity demand
Reference foundation · Category 2
Provider:model auto-router and intelligence proxy.
Inspect the payload, compute the task requirement, filter by Standard, Private, or ZDR, then minimize cost across models capable of clearing the task contract.
Prompt intent classifier
Measure token length, structure, code density, modality, reasoning depth, and requested output before dispatch.
Real-time benchmark index
Maintain a versioned cost-per-capability matrix using provider prices, internal evaluations, and external benchmark references.
Cascade + verification
Try the lowest-cost eligible route first, then fall back when schema, evidence, confidence, or business-rule checks fail.
Context + cache pruning
Remove redundant prompt material, reuse content-addressed specialist outputs, and exploit provider caching where policy allows.
Payload inspection
Context length · code/math ratio · modality · reasoning tags · account privacy filter
Intelligence score
Reference concept: IQ 25 simple text · IQ 68 structured extraction · IQ 94 multi-step logic
Matrix optimization
Require model score ≥ task score, then minimize verified cost within the privacy-filtered route set.
CONCEPT RETAINED · BENCHMARK REQUIRED
The source proposes sub-12ms routing, up to 50% context/cache reduction, and up to 90% savings from capability matching. These remain visible targets until measured—not discarded claims.
AGGREGATION + PROXY
One unified portal, one governed execution gateway.
ATOMS combines an OpenRouter-like provider catalog with Zapier-like workflow automation, then differentiates through privacy filters, capability scoring, verification, and a transparent savings ledger.
Productized aggregation
A single balance can cover public and private routes. Provider differences are normalized into one request, response, usage, and audit schema.
- Unified billing
- 100+ route target
- Normalized structured output
Productized proxy
The gateway applies routing, token accounting, BYOK, fallback, and retention rules without forcing each workflow to own provider-specific logic.
- Provider failover
- ZDR enforcement
- Low-latency policy decisions
REFERENCE FOUNDATION · CATEGORY 3
The service that shows how it pays for itself.
Every verified run should expose baseline cost, optimized cost, client savings, ATOMS yield, and fallback history across daily, weekly, monthly, workflow, and account views.
Reference scenario: 92% below a default flagship dispatch.
1,420 workflow steps across 18 active worker threads.
Reference claim: multimodal overhead removed from 84% of text-only work.
$49 SaaS fee · $1,371 reference net value generated.
Privacy-safe execution telemetry
Measure latency, token yield, schema passes, error codes, and outcome status without retaining raw Private or ZDR prompts.
Cost vs. worth
Compare actual cost and verified quality with the assigned task score and historical route performance.
Autonomous re-tuning
Trial a cheaper eligible model, graduate it after repeated passes, and preserve instant rollback.
ACCOUNT PROFILEMONTHLY TOKEN SPENDREFERENCE MARGINRATIONALE
Starter / MicroUnder $1003.0%
Cover fixed processing overhead.
Standard Team$100–$2,0002.0%
Unlimited members with a shared 25-session pool.
Enterprise Organization$2,000+1.0%
Nested budgets and contracted capacity.
Reference foundation · Category 4
Asset-light first. Owned intelligence when the flywheel earns it.
Use provider infrastructure to prove demand, subscription economics, workflow retention, and the training/evaluation corpus before taking on GPU ownership.
Asset-light intelligence proxy
- Near-zero initial infrastructure CapEx concept
- Instant adaptability as stronger models launch
- Subscription margin plus verified routing yield
Owned compute and ATOMS inference
- Deploy only after workload and runway gates
- Capture more inference margin on qualified tasks
- Maintain external-model rollback and evaluation
REFERENCE SUPERSET
Retained concept inventory
Items stay visible until deliberately validated, revised, or pruned.
One portal, one credit balance, normalized JSON, 100+ public and private model routes.
Low-latency execution gateway, token accounting, failover routing, BYOK, and ZDR enforcement.
Latency, token yield, schema pass rates, and error codes without raw Private or ZDR prompt storage.
Promote cheaper routes after repeated task-contract passes; keep rollback and frontier fallback.
ARR = 12 × [(users × ARPU) + (token volume × routing margin)].
Reference mechanics: base margin below 32k context; +0.25% for heavy context or extended thinking.
Reference mechanics: 1–25 standard threads; +0.50% for bursts above 100 threads.
Start with provider infrastructure and software margins; add owned compute only after demand evidence.
Retain the atoms.ai acquisition assessment as a tracked concept requiring a refreshed market check.
REFERENCE LAUNCH FINANCIALS~$5K float
Near-zero incremental infrastructure concept using existing ATOMS systems.
PROFITABILITY CONCEPTDay-one contribution
No net-new fixed GPU overhead during the proxy phase.
YEAR-ONE TARGET$1.85M–$2.5M ARR
Source assumption: 3,000 subscribers at $49 ARPU plus $350 average routed spend.
All figures in this restored reference layer are internal scenario assumptions carried forward from the supplied HTML. They are not audited forecasts, live prices, or verified performance claims.
ATOMS LLM foundation · model rights
Proprietary, open weight, and open source.
This taxonomy belongs in the owned-model roadmap, not as the customer's primary routing filter. ATOMS treats model provenance and license terms as an internal eligibility gate after Standard, Private, or ZDR is selected.
Proprietary
ClosedThe provider controls weights, training assets, and deployment rights. Customers buy access, not ownership.
Open weight
Inspect the licenseWeights are downloadable, but code, training data, or commercial rights may still be restricted.
Open source
Reproducible rightsCode and rights are broadly available under an approved license. A model can be open weight without meeting this bar.
GLM‑5.2MIT weights
Gemma 4 31BApache 2.0 weights
ATOMS studentRights-cleared derivatives
GLM supplies high-capacity teaching; Gemma supplies a practical, tunable base.
Payload intelligence
Do not pay a frontier model to inspect every pixel twice.
ATOMS analyzes each modality with a specialist, stores reusable embeddings and evidence pointers, and assembles a compact context package for the least-cost reasoner that clears the task-quality floor.
The source remains immutable and addressable for audit.
Each modality is converted once into task-relevant evidence.
Retrieval selects only the evidence required for this question.
Raw media is escalated only when confidence or task type requires it.
Where context vectors can be better
Repeated retrieval, classification, deduplication, and semantic matching over large media collections.
Where raw media still winsPixel-level defects, exact spatial relationships, dense OCR, forensics, and evidence that fails the confidence gate.
Compounding architecture
Every verified payload improves the owned layer.
Govern
Consent, retention, region, and license rules define the eligible pool.
Route
Task-specific evaluations select the lowest-cost model above the quality floor.
Verify
Schemas, graders, and business outcomes decide whether a result is usable.
Graduate
Qualified workloads move to an ATOMS post-trained model with automatic fallback.
24-month roadmap
Four gates. One owned outcome.
Each stage has a measurable graduation condition. Capital is deployed only after product evidence and data rights are proven.
1
0–6 months
Prove the control plane
Ship policy-aware routing, task evaluations, unified metering, and a verifiable savings ledger.
2
7–12 months
Earn the dataset
Capture consented outcomes—not raw prompts by default—and curate reusable enterprise task traces.
3
13–18 months
Post-train ATOMS
Fine-tune and distill open-weight foundations against ATOMS middleware evaluations and safety gates.
4
19–24 months
Launch the owned tier
Graduate qualified workloads to an internally recognized ATOMS model with rollback and external-model fallback.
Enterprise value
The account gets more valuable as workflows accumulate.
Lifetime valueRecurring gross profit × retention × expansion
The strategic upside is migration: once ATOMS owns policy, evaluation, and workflow telemetry, a qualified task can move from third-party inference to the owned model without changing the enterprise integration.
See the thesis operate
Route a real enterprise workload.
Change the privacy mode and operating priority. ATOMS will explain every eligible—and rejected—model.