What Nadir is, how it routes your prompts, what it costs, how the design-partner program works, and how your data is handled.
Questions answered on this page
What is Nadir?
Nadir is an LLM router. Nadir sends each prompt to the cheapest model that can handle it. It can run beside your existing gateway as a decision-only API, which calls no model provider, or execute requests through an OpenAI compatible managed proxy. Each decision receipt records the model, reasoning effort, cache, context, policy, and economics, and can be joined to measured usage and customer outcomes before a policy is enforced. The default adoption path is shadow mode first.
Do I need to change my code?
For the managed proxy it is a two-line change: point your base URL at api.getnadir.com and set model to auto. The decision API needs no change to your request path. You call it beside the gateway you already run, which is how a shadow-mode evaluation works.
What is a verifier-gated cascade?
A verifier-gated cascade first calls a leaner model, scores its complete answer, and may call a stronger model when the answer fails a configured threshold. In Nadir this is an optional managed-proxy policy for complete non-streaming responses. Verification errors fail open, streaming bypasses post-generation verification, and measured quality is determined by customer outcomes, not by verifier acceptance alone.
How is Nadir different from Not Diamond, OpenRouter, or Portkey?
Nadir is designed to run beside an existing gateway rather than replace it. Its decision receipt covers model, reasoning effort, cache, context, governance policy, and advisory economics, then joins measured execution and customer outcomes. That gives an auditable path from shadow observation to reviewed enforcement. Each comparison, with every competitor claim cited from that vendor's own docs and date-stamped, is at getnadir.com/compare/.
Is Nadir free?
The NadirClaw self-hosted core is source-available under the PolyForm Noncommercial License and free for noncommercial use. Nadir Route's hosted execution plan has no base fee and charges a variable fee only on measured savings from requests Nadir executed. The Decision API is priced as a pilot, then flat or usage-based. Decision-only and shadow-mode projected savings are advisory and are never billed as delivered savings.
How are Nadir and NadirClaw related?
NadirClaw is the self-hosted core of Nadir: a local CLI and server that routes each prompt across four model tiers on your own hardware, with your own provider keys, no account, and local-only storage. It is source-available under the PolyForm Noncommercial License, not open source. Nadir is the hosted product built on that core. It adds the decision-only API, the managed OpenAI compatible proxy, decision receipts, savings accounting, organization observability, governance, and optional post-generation verification. Neither product publishes a universal savings rate.
What LLM providers does Nadir support?
Nadir supports Anthropic (Claude), OpenAI (GPT), and Google (Gemini) models out of the box. You can configure custom model tiers and routing rules for any OpenAI compatible provider, and BYOK (bring your own keys) is supported on every tier.
Does routing to cheaper models hurt quality?
Nadir does not publish a universal quality-preserved percentage for production traffic. Quality is customer-specific and is reported only for outcome-labelled requests against a declared baseline. Shadow mode lets each team measure coverage and quality on its own traffic before enabling enforcement.
What are Nadir's benchmark results?
Nadir reports each benchmark with its scope. On checkable code, run-check-escalate solved 392 of 395 common HumanEval and MBPP problems (99.2%), graded by running the canonical tests; that applies only to tasks with runnable deterministic tests. A reference-assisted RouterBench evaluation on 11,420 held-out triples produced a 60% lower projected cost and about 98% retained quality versus always-Opus at one research threshold. That experiment gave the verifier the expensive-model reference answer, so it is a research ceiling, not the deployed reference-free path, a customer forecast, or a production guarantee.
What is the verifier's AUROC and calibration?
The AUROC 0.961 and ECE 0.016 figures belong to a reference-assisted RouterBench research evaluation on 11,420 held-out triples. They are a research ceiling and do not describe Nadir's deployed reference-free verifier. Production quality is reported only from customer outcome-labelled traffic.
How does the Nadir design-partner program work?
A team picks one of three rungs by risk appetite, and every rung ends in a savings receipt on that team's own traffic. Rung 0, Shadow: one advisory decision call beside live traffic, nothing in the request path changes, and the receipt is projected, worded "would have saved". Rung 1, Hosted trial: a two-line swap on a production slice, and the receipt is realized. Rung 2, On-prem POC: a compose bundle in your VPC for a supervised six-week window, the receipt is realized and computed on your infrastructure, and no prompt, response, or usage reaches Nadir. Projected and realized receipts are never blended. Apply at getnadir.com/contact/?reason=design-partner.
How does intelligent routing work?
When you send a request to Nadir, it goes through a multi-step pipeline before reaching a provider. Complexity analysis: your prompt is analyzed by a lightweight ML model (a ~5 MB neural network) that scores its difficulty on factors like reasoning depth, domain specificity, and output structure. Tier assignment: based on the score, the request is classified into a tier (e.g. simple, moderate, complex) that maps to a class of models. Model selection: within the tier, Nadir picks the best available model considering provider health, latency, and your configured preferences. Fallback: if the selected model is unavailable, Nadir walks a per-tier fallback chain so your request still succeeds. The result: simple prompts go to fast, cheap models while complex ones get routed to frontier models, automatically, with no code changes on your side.
What is BYOK vs Hosted mode?
Nadir supports two operating modes. BYOK (Bring Your Own Keys): you add your own provider API keys (OpenAI, Anthropic, Google, etc.) in the Integrations page. Nadir routes requests using your keys, so you pay providers at your own negotiated rate. Nadir adds no fee. Hosted mode: Nadir provides the API keys so you do not need any provider accounts. Credits for Nadir-managed keys draw down at provider list price. A 5.5% platform fee (8% on Business) is added when you buy them. This is the fastest way to get started. You can switch modes or use both at any time from the Integrations page.
How are savings calculated?
Savings are measured by comparing what you actually paid against what the request would have cost on the benchmark model you declare, set on the API key or on your profile, repriced over the same measured token counts. What you actually paid includes every provider call Nadir made for the request, including any escalation. For example, if your declared benchmark is GPT-4o and Nadir routes a simple question to GPT-4o-mini at $0.002 instead of $0.03, your saving on that request is $0.028. The Savings page in your dashboard shows these numbers aggregated over time.
How is Nadir Route priced?
Nadir's only fee is a platform fee on credits for Nadir-managed keys. Your own keys and routing savings carry none. Free: the self-hosted core, NadirClaw, is free for noncommercial use under the source-available PolyForm Noncommercial license. Pay as you go: credits for Nadir-managed keys draw down at provider list price. A 5.5% platform fee is added when you buy them, so $1,000 of credits costs $1,055. No contract, no minimum. Business: the same with an 8% platform fee, plus organizations and a signed DPA on request. Enterprise: negotiated fee discounts and invoicing. Business and Enterprise can also run on-prem in your VPC, on a yearly licence. Your own keys, on every plan: your provider bills you directly and Nadir adds no fee. Routing savings carry no fee: what routing saves is yours. There is no base fee. Nadir Lens, the org observability product, meters separately per org member at $4.99 per member a month, and is free for every org until Lens reaches general availability.
How do I add my provider API keys?
Go to the Provider Keys page under Manage in the sidebar. Click Add Provider, select your provider, and paste your API key. Keys are encrypted at rest and never logged. You can add, rotate, or remove keys at any time.
What happens if my provider key fails?
Nadir maintains a per-tier fallback chain. If the primary model returns an authentication error, rate limit, or timeout, Nadir automatically retries with the next model in the chain. The provider health monitor tracks rolling success rates and deprioritizes unhealthy providers in future routing decisions. You can view provider health status on the Dashboard and configure fallback preferences in Settings.
How does Context Optimize work?
Context Optimize reduces token usage and cost by trimming or restructuring the context sent to the model. It is off by default. Safe mode is deterministic, with no model calls. It removes system prompt text repeated in later messages, replaces repeated tool schemas with a short reference, minifies JSON, and collapses whitespace outside code fences while keeping indentation. Once a conversation passes 40 user turns the middle turns are trimmed, so safe mode is not strictly lossless. Aggressive mode adds semantic deduplication: a message that nearly repeats an earlier one is replaced with a short reference and the phrases that changed. This can yield larger savings but may occasionally remove relevant context. You can turn Context Optimize on in Settings, per API key, or per request with the layers field, for example "layers": {"optimize": "safe"}.
Can I set a spending limit?
Yes. Navigate to Settings to configure a monthly budget cap. Once you reach the limit, Nadir rejects new requests with a clear error message until the next billing cycle, or until you raise the cap. You can also set per-key limits from the API Keys page.
How do I cancel my subscription?
Go to the Billing page and click Cancel Subscription. Your access continues until the end of the current billing period. After cancellation, your API keys stop working but your data (analytics, logs) remains available for 30 days in case you decide to reactivate.
Is my data secure?
Security is a core design principle. Row Level Security (RLS): every database table uses Supabase RLS policies so users can only access their own data. Encryption: provider keys are encrypted at rest, and all traffic is TLS-encrypted in transit. Prompt storage you control: by default Nadir stores prompt and response content so your dashboard and logs can show the traffic back to you; turn prompt storage off in Settings and Nadir writes a SHA-256 hash in place of the prompt, drops the response, and keeps only metadata (token counts, model used, latency, cost). API key hashing: your Nadir API keys are stored as SHA-256 hashes, and the raw key is shown only once at creation time.
What Nadir is
Nadir is an LLM router. Nadir sends each prompt to the cheapest model
that can handle it. A small decision model
reads each prompt against your model menu and picks the lowest-cost model
it expects to handle it. When the decision model is uncertain or
unavailable, the request goes to the strongest model on the menu. The
routing step makes no call to your model provider.
Nadir runs two ways. The decision API returns a model, reasoning-effort,
cache, context, and policy recommendation without calling a model
provider, beside the gateway you already run. That is how a shadow-mode
evaluation works, and its projected savings stay advisory. The OpenAI
compatible managed proxy executes the route, and migration is a two-line
change: point the base URL at api.getnadir.com and set model to auto. On
that path an optional verifier can score a complete non-streaming answer
and escalate to a stronger model when it misses the configured bar.
Streaming bypasses post-generation verification. BYOK is supported on
every tier.
For coding agents, Nadir Auto is a terminal launcher for Claude Code or
Codex that asks Nadir for the main-session model before each turn, with
your configured model as the fallback and cost ceiling. Delegation
integrations for Claude Code, Codex, or Cursor instead recommend a model
tier for subagent work, and the agent decides; the Claude plugin can also
move a main-thread turn to Haiku when Nadir's cache check prices it
cheaper, unless its switch_main_model option is off. Either way inference runs
on your own provider account, so Nadir holds no provider keys on that path.
What the numbers are, and what they are not
Nadir publishes each evaluation with its scope. On checkable code,
run-check-escalate solved 392 of 395 common HumanEval and MBPP problems
(99.2%), graded by running the canonical tests; that applies only to tasks
with runnable deterministic tests. Nadir-Tumbler posts an arena_score of
72.3 on RouterArena's public scorer, 5th of 23 routers, which measures the
routing decision on RouterArena's own model pool. A reference-assisted
RouterBench evaluation over 11,420 held-out triples produced a 60% lower
projected cost than always-Opus with about 98% retained quality and a 1.7%
catastrophic-route rate. That experiment gave the verifier the
expensive-model reference answer, which production does not have, so it
is a research ceiling and not the deployed path.
None of these is a production guarantee, a universal savings rate, or a
forecast for any particular workload. Customer savings are reported from
measured execution against a declared baseline, and customer quality only
from outcome-labelled traffic. Projected savings and realized savings are
separate artifacts and are never blended.
Design-partner program
Three rungs, picked by risk appetite. Rung 0 Shadow runs advisory decision
calls alongside live traffic and returns a projected receipt, with nothing
in the request path changed. Rung 1 Hosted is the two-line swap on a
production slice and returns a realized receipt. Rung 2 On-prem is a
supervised six-week proof of concept inside the partner's VPC, where no
prompt, response, or usage reaches Nadir. The commitments are the same at
every rung. Apply for a rung directly:
Rung 0 Shadow,
Rung 1 Hosted, or
Rung 2 On-prem.
Not sure which fits? Start at
getnadir.com/contact/?reason=design-partner.
Licensing
NadirClaw is the self-hosted core, source-available under the PolyForm
Noncommercial License. Source-available is the correct label; NadirClaw is
not open source. Nadir Route's hosted plan has no base fee and charges a
variable fee only on measured savings from requests Nadir executed.