How this works

Public cohort data → Full methodology → Audit log →

Legal: Terms of Service · Privacy Policy · Data Processing Agreement (request a signed copy at legal@ordica.ai)

0 characters Ctrl+Enter to analyze
Models:
Samples:

Your prompt is processed in memory only — never written to disk, never sent to a model for inference. We log content-free metadata for abuse prevention. View audit policy →

Ordica reduces your LLM spend by 7–50%. The exact number depends on your prompts, which we have not yet seen.

Conversation history: 35.5% median. Long instructions: 28% median. RAG retrieval: 42.8–55.3% (median 49.8%). Mixed production traffic: 7–10%. Blind-tested across 200 prompts × 4 providers — ranges published by workload type, not averaged into a single marketing number.

# Private beta. No new SDK. No code rewrite.
# Approved partners swap one env var:

export OPENAI_BASE_URL="https://api.ordica.ai"
# — or —
export ANTHROPIC_BASE_URL="https://api.ordica.ai"

# Your OpenAI or Anthropic SDK keeps working. You keep your key.
# Request access for BYOK onboarding →
VERIFIED SAVINGS BY WORKLOAD
Your savings depend on what you send us.
Here is the honest range.
◎

Conversation history

33% median

Multi-turn chat histories carry the most redundancy. When we see a conversation that has grown past 2K tokens, the compression gain is the highest of any workload class we measure.

▷

Instruction-heavy prompts

28% median

Long system prompts, detailed role definitions, and few-shot exemplars compress well. Typical result for production AI applications with heavy prompt scaffolding.

◇

Mixed structured traffic

7–10%

When your prompts are already dense — short JSON payloads, minimal instructions, tight function-calling — savings are modest. We publish this. When compression won't help your prompt, we pass it through at 0% — you're never charged a fee on a pass-through.

⬡

RAG retrieval context

32–38% depending on provider (median ~35%)

Blind-judged equivalence 4.35 / 5.0 across 175 validations

Retrieved document blocks show the highest savings of any workload class we measure. Savings validated on 50 prompts × 4 providers; quality validated on 175 prompts (OpenAI 50, Claude 50, Grok 50, Gemini 25 — see methodology for Gemini sample note).

Token savings measured across 200 unique prompts (one deterministic savings number per prompt) plus 775 blind quality judgments across all four cohorts — 50 prompts each for OpenAI, Claude, and Grok, with Gemini at 25 pending an API parser fix. Response quality was separately blind-judged on a 1–5 scale by a separate AI; per-provider means are 4.38 on OpenAI, 3.96 on Claude, 3.78 on Gemini, 3.76 on Grok. Three independent judges (GPT-4o, Gemini, and Grok-3) separately cross-validated the same corpus — aggregate corrected mean: 3.88 / 5.0 with 74% exact agreement between the two most consistent judges. RAG-cohort means: OpenAI 4.22, Claude 4.22, Gemini 4.40, Grok 4.60 — cross-provider 4.35 / 5.0 on 175 RAG validations. Full methodology →

How it works
Three steps. Sixty seconds.
1

Point

Set one environment variable
OPENAI_BASE_URL or ANTHROPIC_BASE_URL
= https://api.ordica.ai

2

Authenticate

Keep your existing provider API key
We forward it to your chosen provider
and don't retain it after the request

3

Save

Compressed when safe, passed through when not.
Quality tracked against blind-judged equivalence scores. Fewer tokens when we can; your original prompt when we can't.

xAI users: the OpenAI SDK already works — same env var, your Grok API key. Gemini users: pass http_options={"base_url":"https://api.ordica.ai/gemini"} to the google-genai client, or contact us for deeper integration support.

Privacy
Prompts pass through.
Nothing sticks.

Your prompts transit the Ordica proxy in-memory, get compressed, and are forwarded to the AI provider you chose. We do not store them, log them, read them, or train on them. The only data retained is anonymous counts: tokens sent, tokens saved, and which provider you used.

Billing runs on those counts alone. Prompt content is not persisted beyond the compression step — our audit logs contain metadata only: token counts, timestamps, billing meters. Retention and sub-processor details are enumerated in our Data Processing Agreement.

Your prompt → Ordica proxy compressed → Your AI provider
Token counts → Ordica telemetry anon
Message content → Our storage never retained
Compatible
Works with the AI you use

Optimized for each provider. Switch between them freely.

GPT-4o
Claude
Gemini
Grok
Validated
Measured, not promised

Every message is a real-world test. A separate judge model scores compressed vs. original responses without knowing which is which.

GPT-4o
4.38
/ 5.0 validated mean
Instruction cohort4.78
Mixed cohort4.62
RAG cohort4.22
Claude
3.96
/ 5.0 validated mean
Instruction cohort4.10
Mixed cohort4.10
RAG cohort4.22
Gemini
3.78
/ 5.0 validated mean
Instruction cohort3.94
Mixed cohort4.34
RAG cohort (n=25)4.40
Grok
3.76
/ 5.0 validated mean
Instruction cohort4.06
Mixed cohort3.58
RAG cohort4.60

✓ Cross-validated by three independent judges (GPT-4o, Gemini, and Grok-3) across all 200 prompts. Inter-judge agreement: 74% exact match, 99% within one point. Cross-judge corrected mean: 3.88 / 5.0. Full results and per-provider breakdown at methodology: cross-judge validation.

200
prompts × 4 providers
4
providers validated
3.88
cross-validated mean / 5.0
3
independent judges

Blind-tested across 200 prompts × 4 providers: a separate AI judged compressed vs. original responses without knowing which was which. Equivalence is scored on a 1–5 scale, where 5 means indistinguishable and 4 means "substantively equivalent — a customer would accept either response." Mean equivalence above 3.5 on this scale indicates the compressed response reliably passes blind substitution. Scores cluster in the 3.5–4.6 range across providers because compression is probabilistic, not lossless — we publish the real distribution rather than round up. All four cohorts (instruction, history, mixed, RAG) are now quality-judged; the RAG cohort shows a cross-provider mean of 4.35 / 5.0 across 175 validations (Gemini ran at n=25 due to a parser fix now resolved). Savings are consistent and repeatable for the same prompt under equivalent configuration. Grok token counts in our benchmark runs used a documented approximation — see methodology for per-model details.
All tests run on current production models. No cherry-picking. Full methodology and the per-prompt judge output: analyze/methodology.html.

Estimated annual savings

Savings are consistent and repeatable for the same prompt under equivalent configuration. Grok token counts in our benchmark runs used a documented approximation — see methodology for per-model details. Figures below use the four-cohort blended median (30%). The calculator above defaults to the single-cohort Conservative figure (10% p25) for a lower-bound estimate — toggle the Blended view to apply this 30% median to your spend. Actual savings depend on workload mix — RAG-heavy prompts save more, mixed-structure prompts save less.

Monthly API spend Conservative (p25, 10%) Median (30%) Optimistic (p75, 40%)
$1,000 /mo $1,200 $3,600 $4,800
$5,000 /mo $6,000 $18,000 $24,000
$25,000 /mo $30,000 $90,000 $120,000
$100,000 /mo $120,000 $360,000 $480,000
Savings rate 10% (Conservative p25) 30% (median) 40% (blended p75)

See your number

Your monthly LLM spend
$ / month
Gross savings
$2,500/ month
$30,000 cut from your annual LLM bill.
Pro tier — you keep 70%
$1,523 / mo
$18,270 / year
Enterprise — you keep 80%
$1,740 / mo
$20,880 / year
Default shown is the Conservative cohort (dense structured traffic, 10% p25) — switch to Blended above to see median or higher-savings projections. The projected savings table uses the blended four-cohort mean (30%, RAG-inclusive); the calculator default is deliberately lower for a cautious estimate. Based on 200 prompts × 4 providers across GPT-4o, Claude, Grok, and Gemini. Savings are consistent and repeatable for the same prompt under equivalent configuration. Grok token counts in our benchmark runs used a documented approximation — see methodology for per-model details. Validated range 7% to 50% by workload (median by cohort). We make money when tokens disappear. We make nothing when they don't. Call it performance-based pricing, or call it the only business model where lying to you is directly unprofitable. See pricing →
Validated
Savings by workload type

Results across 200 unique prompts (one deterministic savings number per prompt, 4 providers verifying each) with 775 blind-judged quality validations across all four cohorts. Gemini RAG validations ran at n=25 instead of 50 because of a parser fix now resolved — the other three providers ran the full 50 RAG validations.

Conversation history: 33% median. Instruction-heavy workloads: 28% median. RAG retrieval context: 32–38% depending on provider (median ~35%) — blind equivalence 4.35 / 5.0 on 175 validations. Dense structured traffic: 7–10%. Savings depend on your workload mix and are consistent and repeatable for the same prompt under equivalent configuration. For Grok, our benchmark runs used a documented approximation — see methodology for per-model details. Quality preservation varies by provider.

Domain Benchmarks
How your document type changes the math

The 7–38% range above reflects mixed production traffic across all workload types. Document-processing prompts — where a single prompt carries a full document — compress further: 44–81% depending on document type, versus 7–38% for mixed workloads. These figures come from a separate domain-specific benchmark: 200 samples per domain, public-domain sources, reproducible at seed=42.

◈

Financial documents

81% mean token reduction

Measured on public SEC EDGAR filings — annual reports, 10-K exhibits, and quarterly disclosures. p25: 76% · p50: 84% · p75: 88%. Financial filings carry dense repetitive structure — boilerplate disclosures, repeated header blocks, and standardized exhibit language — which is why reduction rates here run significantly higher than general production traffic.

◉

Regulatory documents

44% mean token reduction

Measured on Federal Register rules and proposed rulemakings. p25: 45% · p50: 46% · p75: 48% · p95: 51%. The p25–p95 central distribution spans only 6 points — most documents compress in a tight band. The arithmetic mean (44%) falls slightly below p25 because a small number of long regulatory documents with lower compressibility pull the average down; the central distribution is the better planning estimate for most workloads.

Reproducible benchmark · seed=42 · 200 samples per domain · public-domain sources (SEC EDGAR, Federal Register). These numbers come from document-processing workloads. General LLM traffic will produce different results. We don't know yours yet. Run the trial. Domain coverage is expanding — financial and regulatory are the first two measured; additional document types are in progress. Methodology →

Pricing
You only pay when we save you money

Your billing dashboard shows how many tokens you sent, how many tokens we compressed them to, and what that saved you in dollars. Your bill is 30% of that dollar savings on Pro, 20% on Enterprise, zero on Free. No compression, no fee.

Free
$0
Real savings. No credit card. No catch.
  • Drop-in proxy endpoint — point your existing SDK at api.ordica.ai
  • GPT-4o, Claude, Gemini, Grok
  • Conversation-history workloads: 33% median
  • Up to 10K requests/month
  • Savings dashboard
  • Community support
  • Best-effort availability — see Terms
What you get: Point your existing OpenAI or Anthropic SDK at Ordica with one environment variable. No new package, no code rewrite. Token savings measured by provider tokenizers across 200 unique prompts (one deterministic savings number per prompt, 4 providers verifying each). Response-quality preservation separately blind-judged on a 1–5 scale by a separate AI — see full methodology.
Most popular
Pro
30% of your measured savings
You keep 70% of every dollar we save you.
  • Same drop-in proxy as Free
  • Full compression engine
  • Instruction-heavy workloads: 28% median
  • 100K requests/day
  • Priority support
  • Full savings analytics
  • Production-grade uptime target — see Terms
Why upgrade: Higher rate limits, the full compression engine, and full analytics. Savings are consistent across providers; quality preservation is where providers differ — see the analyzer for per-provider quality scores.
Enterprise
20% of your measured savings
Custom-tuned for your prompts. Maximum savings.
  • Everything in Pro
  • Custom optimization profile
  • On-prem / air-gapped SDK deployment — roadmap; available for qualified prospects under custom engagement. Contact sales for current availability and attestation requirements.
  • Direct access to the founder
  • Custom SLA under enterprise contract
  • Unlimited requests
  • Annual commitment, terms negotiated per contract
  • On-prem billing uses signed usage attestations from your own telemetry
Why 20% not 30%: At enterprise scale, your usage helps us improve optimization for all customers. The lower rate is earned, not negotiated. On-prem SDK deployment is a roadmap feature offered under custom enterprise engagement; availability, timeline, and the specific attestation mechanism are defined per contract. When Ordica cannot observe traffic directly, billing runs off signed usage attestations your team generates from your own telemetry — contract defines the exact format and cadence.

Pricing is structured so your Pro or Enterprise fee is a fraction of your measured per-period savings — 30% on Pro, 20% on Enterprise, zero on Free. If a billing period shows no measured savings, there is no fee. The exact guarantee, edge cases, and dispute process are set out in our Terms.

Automatic monthly renewal — please read.
By starting a paid plan, you authorize Ordica LLC to charge your payment method on a recurring monthly basis until you cancel. Your subscription automatically renews at the then-current price at the end of each monthly term. You may cancel at any time from your account dashboard or by emailing support@ordica.ai; cancellation takes effect at the end of the current billing period and stops further charges. Refund and cancellation terms are set out in our Terms.
Questions
Frequently asked

No. Your messages pass through the proxy to reach the AI provider. Our audit and billing logs contain metadata only — token counts, timestamps, provider, and billing meters. Prompt and response content is not persisted beyond the compression step. Retention windows, sub-processors, and data-subject rights are enumerated in our Data Processing Agreement.

Sometimes. We ran 200 prompts across 4 providers — 800 blind evaluations total. Compressed outputs scored 3.88/5.0 on average. We recommend running your own evaluation before committing. Benchmark protocol →

The compression adds a few milliseconds. The provider's response time dominates. Shorter prompts can produce faster provider responses — fewer tokens to process.

Ordica's compression is deterministic on our tested corpus — same input, byte-identical output. Cache keys stay stable. In our test harness, a compressed prompt produced a full read-hit on the second call (Anthropic cache_read_input_tokens equal to cache_creation_input_tokens). Validate your own prompt shapes. cache_control markers are left untouched. Provider-side cache hits are unaffected.

If compression confidence is insufficient, the prompt passes through unchanged. You lose the savings on that request. You do not get a degraded response. Worst case is zero savings, not worse quality.

Billing runs on token counts. For every request: tokens you sent, tokens we forwarded. The difference is your savings. Your bill is a fixed percentage of that. Stripe receipts show the metered total. Your account shows per-request counts. Skipped compression contributes zero. Enterprise: full audit export on request.

GPT-4o and Claude handle general workloads. Gemini has the largest native context window. Grok is fast on short prompts. Run the analyzer. It will tell you which one scores best on your specific prompt type.

Yes. The Free tier gives you the drop-in proxy endpoint — point your existing OpenAI or Anthropic SDK at Ordica with one environment variable — 10,000 requests per month, and a savings dashboard. No credit card required. You get the same compression engine as paid tiers.

We charge a percentage of your measured savings — 30% on Pro, 20% on Enterprise. If compression saves you $100, you keep $70 (Pro) or $80 (Enterprise). If it saves you nothing, you pay nothing. Your dashboard shows every dollar in real time. No flat fees, no minimums.

Your provider API key transits the Ordica proxy in-memory. We use it to authenticate with the provider you chose — OpenAI, Anthropic, xAI, or Google — on your behalf for that single request, then discard it when the request completes. We do not retain it, log it, store it in a database, or share it. The key is used only to reach the provider endpoint you designated; it never leaves that path.

While your request is in flight, the key is held in process memory on our proxy layer. Ordica is not a secrets vault or KMS. You hold the key in your code or secret manager, and you can rotate it through your provider's dashboard to revoke access.

Routing your own API key through a middleware layer is a standard pattern — every major provider (OpenAI, Anthropic, xAI, Google) builds an API ecosystem around SDK wrappers, gateways, observability proxies, and caching layers. Ordica fits within that pattern: you remain the authorized account holder, your requests go only to providers you designate, and we do not resell, pool, rent, or share API access across customers. We do not retain prompt or response content after each request completes.

Ordica fits within what each provider permits. What providers prohibit — multi-tenant key pooling, API-key resale, unauthorized account sharing — is not what we do. Provider Terms change. Review yours. For enterprise documentation — DPAs, subprocessor list, data-flow diagrams — contact us.

We route to any OpenAI-compatible endpoint. If one provider changes pricing or access, traffic moves to another. Your integration does not change. We do not depend on the continued goodwill of any single provider.

We do not cache responses. We do not match prompts. Every request goes to the model and comes back fresh. We reduce what you send — we do not decide what counts as equivalent. There are no stored answers to return incorrectly.

Every request produces two token counts: what you sent, and what we forwarded. The difference is your savings. Both are logged. You can audit them.

Ready to compress without compromise?

Free tier is open globally. Pro and Enterprise require a US billing address.

Aurora Ordica AI Support
Email
First name
Question or message