Skip to content

Nautili.aivsPortkey

Both are LLM gateways. Nautili is built around spend per developer and budgets that hold, and the table below is clear about where Portkey fits better.

Nautili.ai vs Portkey at a glance

Both put routing, fallbacks, caching and filters behind one endpoint. Where they part ways is which plan gets you budgets, and whether your guardrails need to read the model's answer as well as the prompt.

Choose Nautili if
  • You want spend broken down by developer, team and service, with budgets on Pay-as-you-go.
  • Claude Code, Codex and Cursor usage should show up under each developer's name.
  • You want to test a cheaper model on live traffic before you switch.
Choose Portkey if
  • Your guardrails have to check model responses, not only prompts.
  • You want an MIT-licensed gateway you can fork and run yourself.
  • Your security team is standardizing on Palo Alto Networks Prisma AIRS.
Both give you
  • One OpenAI-compatible endpoint in front of many providers.
  • Fallbacks across models and providers, and response caching.
  • A choice between hosted and self-hosted.

Feature comparison

Routing, spend, governance and deployment, row by row, starting with where the two differ.

Where they differ · 7
BudgetsGovernance
Nautili.aiPay-as-you-go

A hard dollar cap for a person, a key or a whole team, resetting daily, weekly, monthly or never. You can add requests-per-minute and tokens-per-minute limits too.

PortkeyEnterprise or select Pro

Cost or token limits on providers and workspaces, starting at $1, resetting weekly, monthly or never. A limit can't be edited after it's set.

Access controlGovernance
Nautili.aiIncluded

Allow and ignore lists for models and providers, with * wildcards. Shadow mode logs what a rule would have blocked before you enforce it. Requests on your own credentials still go through; on Nautili-issued keys a would-deny still fails.

PortkeyRate limits on Enterprise

Workspaces and API keys on every plan. Rate limit and usage policies that match on key, provider or model, including wildcard model names, are listed for Enterprise self-hosting only.

Content rulesGovernance
Nautili.aiPrompts only

Built-in detectors for emails, phone numbers, SSNs, card numbers, names, addresses and prompt injection, plus your own custom patterns. A rule can flag, redact or block. Model responses aren't checked.

PortkeyPrompts and responses

A large library of guardrails runs inside the gateway, on requests and on responses.

EvalsQuality
Nautili.aiIncluded

Shadow trials on live traffic, where the candidate answers beside your current model and users never see it, plus replays of past traffic, experiments on your own datasets and a side-by-side compare of up to six models. A person always makes the final call on switching.

PortkeyCanary, no shadow trials

Canary and A/B splits send a share of live traffic to a candidate model, so some users get its answers. There are no shadow trials, where users never see the candidate, and no promotion workflow.

CachingPerformance
Nautili.aiIncluded

Exact-match caching first, then semantic caching where your settings let content leave the deployment. Every response tells you whether it came from the cache.

PortkeySemantic on Enterprise

Exact-match caching on every plan. Semantic caching needs a vector database and is on select Enterprise plans.

DeploymentPlatform
Nautili.aiHosted or self-hosted

Hosted by Nautili, or run it yourself with Helm on Kubernetes or Docker Compose. A Shared Gateway is already set up for your first test on the Demo tier.

PortkeyCloud or open source

Portkey Cloud, or the MIT-licensed gateway on npx, Docker, Node.js or Cloudflare. The enterprise version can also be self-hosted.

PricingPricing
Nautili.aiShare of spend

A Demo tier to start, then 4% of inference spend once you add credits, with tokens at provider list price. There's no fee on the first $50K a month that runs on your own keys.

PortkeyPlan plus logs

Developer is free for 10k logs a month. Production is $49 a month for 100k logs, then $9 per extra 100k requests, up to 3M requests. Enterprise is priced on request.

Where they match · 2
Routing and fallbacksRouting
Nautili.aiIncluded

A routing profile for a team, person or key sets the default model, up to four fallbacks and how providers are ranked: by cost, latency or throughput. If a request names a model, that model is what it gets.

PortkeyIncluded

Gateway configs cover fallbacks, retries, timeouts and load balancing across more than 1,600 listed models.

Spend attributionSpend
Nautili.aiIncluded

Spend is split by person, team, department, service and seat, using a team tag on each request or the key owner. Each number is marked as reported by the provider or measured by the gateway.

PortkeyIncluded

Analytics shows cost per request, and usage policies can group spend by request metadata such as user.

Pricing compared

You pay for the two in different ways. Nautili takes a percentage of inference spend and passes tokens through at list price, while Portkey charges a monthly plan sized by how many logs you record, and your providers bill you separately.

Nautili.aiNautili pricing
4%platform fee on inference spend
  • Demo. Free starter credit to try Nautili, no card
  • Pay-as-you-go. 4% fee, tokens at provider list price
  • Your own keys. No fee on the first $50K a month, then 4%
  • Enterprise. Volume pricing, SSO/SAML and SCIM

Pay-as-you-go includes budgets, prompt caching and per-team attribution.

Try it free
$49a month on Production
  • Open source. Free under the MIT license, self-hosted
  • Developer. $0 for 10k recorded logs a month
  • Production. $49 for 100k logs, $9 per extra 100k requests, up to 3M
  • Enterprise. Priced on request

Budget limits are listed for Enterprise and select Pro customers.

When Nautili.ai is the better choice

Nautili is built for the person who owns the AI bill and has to explain it, team by team:

  1. Budgets, rate limits and model access rules come with Pay-as-you-go, and shadow mode logs what a rule would have blocked before you enforce it.
  2. Spend is broken out by People, Teams, Departments, Services and Seats, and each number tells you where it came from.
  3. Trials score a candidate model on live traffic, and nothing changes until a person approves it.
  4. Claude Code, Codex and Cursor still run on their seats or your keys, and you see usage for each developer.

When Portkey is the better choice

Portkey started as an open-source project and now sits inside a larger security platform. Pick it when:

  1. Your guardrails need to inspect model responses as well as prompts, from a large library of guardrails.
  2. You want to read, fork and host the gateway code yourself under the MIT license.
  3. You'd rather buy AI runtime security and the gateway from the same vendor.
  4. You already pay providers directly, well over $50K a month, and prefer a flat plan priced by log volume.

How to switch from Portkey to Nautili.ai

You don't need a cutover date. Move one app or team at a time while Portkey keeps serving the rest.

  1. 01Connect

    Start on the Shared Gateway

    Every team has one from the Demo tier onward, so the first test needs no install. Change the base URL and the key.

  2. 02Tag

    Move your metadata to Nautili tags

    Tag each request with a user, team or project. Nautili strips the tags before the request reaches the provider.

  3. 03Shadow

    Rebuild your limits in shadow

    Hosted gateways start in shadow mode, which logs what would have been blocked. Requests on your own credentials still go through; on Nautili-issued keys a would-deny still fails.

  4. 04Enforce

    Turn on enforcement team by team

    Once the shadow log looks right, switch that team to enforce and move on to the next one.

Frequently asked questions

What teams usually ask before they switch.

Read the docs
  • What is the difference between Nautili.ai and Portkey?

    Both are LLM gateways with routing, fallbacks, caching and guardrails. Nautili centres on spend per developer and per team, and includes budgets and access rules on Pay-as-you-go. Portkey offers an MIT-licensed gateway with guardrails on prompts and responses, and is now owned by Palo Alto Networks.

  • Can I switch from Portkey to Nautili.ai without changing app code?

    If your app uses the OpenAI, Anthropic or Vercel AI SDK, you change the base URL and the key. Portkey model names start with your own provider name (@your-provider-name/model), so you'll rename them to Nautili's provider/model names, such as anthropic/claude-sonnet-4.5. Your Portkey metadata maps to Nautili's user and team tags.

  • Does Nautili.ai attribute spend per developer and team?

    Yes. Spend goes to People, Teams, Departments, Services and Seats based on workload rules, the key owner and request headers, and headers win when they disagree. Spend is measured by your gateway. When the provider's own figure for managed spend differs after reconciliation, the Spend page shows it beside the gateway's.

  • When is Portkey the better fit?

    Portkey makes sense if your guardrails need to check model responses, or if you want to self-host an MIT-licensed gateway. It's also a natural choice for companies standardizing on Palo Alto Networks Prisma AIRS.

  • How does pricing compare?

    Nautili charges 4% of inference spend, passes tokens through at list price, and takes no fee on the first $50K a month on your own keys. Portkey has a free Developer plan, a $49 a month Production plan with 100k recorded logs, and Enterprise pricing on request.

All comparisons

Point one URL at Nautili

Try the Demo tier on the Shared Gateway and see spend by team before you move anything off Portkey.