Lock Down Your Data.
Cut Token Spend
by 46%, on average.
68% average token reduction
Opto with its optional cost governor, compared with neither. Average of 30 simulated workdays for a 50-person company.
Stop streaming confidential files and raw database rows to public LLM APIs. Opto by Modulus provides an in-memory execution air-gap: transformations, SQL joins, aggregations, and PII redaction execute locally in private RAM—emitting only compact answers.
Sandbox access is granted after a 15-minute walkthrough call — no self-serve signup yet.
Opto in 60 Seconds
46%
Average Spend Reduction
68%
Average Token Reduction
355
Real Measured Task Runs, Underlying Sample
Simple
to Deploy: a Small Local App, Windows or Linux
46% spend and 68% token reduction are the averages of 30 simulated workdays, Opto with the cost governor against neither. They are not guarantees — actual savings depend on your query mix. See the full methodology & fact sheet →
30 Simulated Workdays, Averaged
A 50-user, 600-minute workday simulated 30 times, with query sizes taken from a frozen campaign of 355 real measured task runs (input tokens, output tokens, dollar cost, per-mode, per-cache-state). The line is the 30-day average and the shaded band covers the middle 80% of days. The chart compares two of the four modeled cases: no Opto and no governor, against Opto with the cost governor.
30 simulated workdays (seeds 1001 to 1030), each 600 minutes, sampled every 5 minutes. Line = average across the 30 days; band = 10th to 90th percentile; "lower" = average of the 30 daily reductions. Rate view = 5-minute rolling average. The same 30 days are replayed through every case. Reproducible from the published run files.
The simulation compares four cases. Opto and the cost governor are separate: Opto works with any model on its own, and the governor is an optional add-on that steers each request toward a frontier or a cheaper model. Averages of the 30 days:
| Case | Daily spend | Spend vs. neither | Tokens vs. neither | Frontier requests that got the frontier model |
|---|---|---|---|---|
| Neither | $92.10 | baseline | baseline | 100% |
| Governor only | $66.30 | 28% lower | 6% higher | 23% |
| Opto only | $58.62 | 36% lower | 71% lower | 100% |
| Opto + governor | $49.27 | 46% lower | 68% lower | 64% |
The governor saves money by serving some requests with a cheaper model; it steers toward a configurable daily spend target and does not block requests. The simulation does not model rate-limit rejections or user retries. Details at methodology.html.
Opto's own effect per query type, measured directly from real task runs on Claude Sonnet 5 (a Claude Haiku 4.5 table is in the methodology), varies by size and cache state:
| Size | Cache | Token Reduction | Cost Reduction |
|---|---|---|---|
| Small | Cold | 80.2% | 79.1% |
| Small | Warm | 81.4% | 71.2% |
| Medium | Cold | 64.9% | 51.3% |
| Medium | Warm | 67.8% | 14.3% |
| Heavy | Cold | 57.5% | 40.9% |
| Heavy | Warm | 71.6% | 46.1% |
Cost reduction narrows on warm-cache queries because Anthropic's own cache discount already lowers the unaided baseline's price — full methodology at methodology.html.
The Sovereign Substrate Advantage
Routing decides which model answers; Opto decides how costly and dangerous the hands are.
Optocoupler Isolation
Operates as a true optical-style air-gap. The model has zero shell access, zero filesystem traversal, and zero raw socket access. It orchestrates strictly through verified, declarative tool contracts.
In-Flight Egress Governance
All outbound context passes through compiled in-memory PII scrubbers (SSNs, emails, API keys) and strict byte/row limiters. Oversized datasets are throttled before leaving server RAM.
Dynamic Execution Circuits
Avoid monolithic tool sprawl. Opto orchestrates client-side tooling like Excel parsers, SQL joins, and aggregators into fast execution paths, entirely in private RAM.
Drop-In Compatible, Windows or Linux
Installs as a native local tool server, no dependency stack to manage. Plugs into Claude Code, Antigravity, Cursor, or any enterprise AI agent architecture.
Optional Cost Governor
Opto saves tokens with any model on its own. The cost governor is an optional add-on that complements it: a feedback controller that runs in a gateway, not in the local install, and steers each request toward a frontier or a cheaper model to follow a daily spend target. In our simulation it adds about ten points of spend savings on top of Opto's, with roughly a third of frontier requests served by a cheaper model.
Small App, Full Control
No large appliance to stand up and manage. Opto deploys as a small local app through your existing corporate tooling and approval policies, the same way any other application does — which means your own security team can independently verify and certify exactly which capabilities are enabled, on infrastructure you already control.