See and govern every dollar of your AI.
Veltrix unifies OpenAI, Anthropic and Google in one panel with budget control, alerts and audit. Start free just seeing what you spend and move up to automatic optimization and governance when it makes sense.
7 days · 350 calls · no card · Gemini included
API key ≠ subscription. The Gateway proxies API keys (OpenAI, Anthropic, etc). If you use ChatGPT Plus / Claude Pro, check the Otimizador plan below.
Every call goes to the cheapest model that can handle it.
Flow illustration · percentages are an example of a common chatbot distribution
One line. Zero refactor. Compatible with any OpenAI-standard SDK.
from openai import OpenAI
client = OpenAI(
api_key="vlx_live_•••",
)
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Hello"}]
)One proxy. Two pillars.
AI FinOps and infrastructure Observability — the two things missing from your stack today, in a single endpoint.
FinOps for AI
Prompt compression + semantic cache + routing to the cheapest model that can handle it. On the Gateway, it runs automatically on every call — without changing a single line of code.
"The budget isn't infinite. If you want real AI efficiency, stick with me."
AI Observability
Unified logs from OpenAI, Anthropic, Google and more — in a single panel. Cost per request, latency per model, cache hit rate. Step out of the black box.
"Running multiple AIs today is like flying a plane in the dark. We bring the clairvoyance."
4 subscriptions. 1 map.
You know when you're about to run out.
Lens captures real usage on every provider through the extension. It shows how much of each plan you've burned, and warns you when one is set to run out before the month ends.
Every profile, its own track.
A three-step ladder. Start free just seeing your spend, move up to Otimizador to test and trim prompts, and reach the Gateway when you want automatic governance. Each step includes the previous one.
These prices cover the Veltrix service only. On the Gateway, the tokens consumed at the providers (OpenAI, Anthropic, etc) are still billed to your own API account with them — separately. Veltrix governs and optimizes that bill, but doesn't replace your account at the providers.
You just want to see how much you spend and consume
You use ChatGPT/Claude in the browser and have no idea how much you consume — Lens shows it.
No card · forever
- Veltrix Lens (Chrome extension) monitoring 8 providers
- See how much you spent and how many messages you used
- Unified dashboard (ChatGPT, Claude, Gemini, Cursor…)
- Cost preview in the Playground: estimate before you run
- Searchable usage history
Rules: Free forever. Visibility only — it doesn't optimize or run prompts for you.
Doesn't include running prompts in the Playground or automatic governance.
No card
You pay for ChatGPT, Claude or Cursor and want to stretch it
You pay for Claude Pro and hit the limit before month-end — the Playground teaches you to trim the prompt and stretch it.
Monthly · cancel anytime
- Everything in Free · Lens
- Full Playground: run prompts and compare models side by side
- Effectiveness analysis: tokens, cost and latency per prompt
- Learn to cut verbose prompts and stretch your monthly limit
- Full, searchable execution history
Rules: R$ 39/mo, monthly, cancel anytime.
It's a toolkit that helps you optimize — not automatic compression nor governance of your agents.
7-day free trial · no card
Your software or agents call the API and need governance
Your fintech runs an agent that calls the API all day and the bill grows out of control — the Gateway governs and cuts it, automatically.
Floor R$ 179/mo + 1.5% of spend under management
- Everything in Otimizador
- Control plane: optimizes and governs the calls for you
- Compression + cache + automatic routing on every request
- Spend limits, multichannel alerts (Slack, Discord, webhook) and optional hard cap per project/agent
- Automatic protection against aggressive compression — never sends a degraded prompt to the provider
- Auditable savings panel (before/after)
Rules: Floor R$ 179/mo + 1.5% of the spend routed through Veltrix.
It doesn't replace your provider accounts — OpenAI/Anthropic/etc tokens are still billed to YOUR own API account with them.
Reply within ≤ 24 business hours
Your whole team, governed in a single place
Your team pays for several AI subscriptions and still runs the API, and nobody has control — Teams brings it all together, per seat.
Min 3 seats · all modules
- Everything in Gateway + Lens
- Lens + Proxy + cache (exact + semantic) + Specs (SDD), per seat
- Consolidated team savings panel
- Cost attribution by provider/model with spend drift alerts
- Governance, limits and SSO per user
- Per-seat billing via AbacatePay
Rules: R$ 219/seat/mo, min 3 seats. + 1.5% of the spend that exceeds the per-seat usage cap.
reply within 1 business day
Otimizador and Gateway cover OpenAI, Anthropic, Google, DeepSeek, Mistral, Groq and Cohere. Lens monitors your usage across 8 providers.
Large team, SSO, MSA contract? Enterprise plan →
Buying that separately costs more than the whole Gateway.
We researched comparable tools on the market. Each one solves ONE part of the problem — and charges for it, alone. Veltrix runs all three levers together, automatically, on every call.
Semantic compression
Rewrites the prompt and cuts redundancy while keeping intent — fewer input tokens on every call.
Exact + semantic cache
A repeated or similar question (cosine ≥ 0.88) comes back instantly, without calling the LLM again.
Dynamic routing
Every call goes to the cheapest model that can handle the task, within the quality floor you set.
All three run together, automatically, on every call — it's not a bolt-on, it's the full Gateway.
Blocks and alerts on spend. Doesn't rewrite prompts, doesn't route to the cheaper model, doesn't cache.
Shows spend on a dashboard. Doesn't optimize anything — and most cover only one provider (Azure OpenAI).
Detects a repeated prompt and leans on the provider's native cache. No dynamic routing between models, no semantic compression of its own.
Org-wide circuit breaker: cuts access when the cap is hit. Doesn't lower the cost per call.
- Semantic prompt compression on every call
- Exact + semantic cache — never pay twice for the same answer
- Dynamic routing to the cheapest model that can handle it
Competitor prices cited as a public market reference — direct research on each tool category's site, subject to change. Not a full feature-by-feature comparison; it's cost-per-single-function context vs. a full proxy. Veltrix Gateway pricing per the table above.
Questions that always come up.
read firstCan I use my Claude Pro / ChatGPT Plus / Gemini Advanced subscription?+
It depends on what you want to do. Your monthly subscription (Pro/Plus/Advanced) is for use via browser/app — that's ONE product. An API key is ANOTHER product, prepaid per token, and it's what the Gateway needs to route programmatic calls. They're separate billings on the same account. If you use the AI in the browser and want to get more within your limit, the Otimizador plan (R$ 39/mo) gives you Playground + effectiveness analysis so you can trim verbose prompts yourself and stretch the subscription. If your software calls the API per token, then the Gateway optimizes and governs automatically.
Can I use it with OpenClaw, Hermes Agent, Cline, Claude Code and other autonomous agents?+
Yes — that's where the Gateway shines most. Agents consume 5-50× more tokens than a human chat (read+edit+verify loops), so the savings in $ become huge. Setup: 1 env var per agent. OpenClaw: `LLM_BASE_URL=https://api.veltrixplatform.com/v1`. Cline: 'OpenAI Compatible' settings pointing to the same URL. Aider: `--openai-api-base`. Claude Code: `ANTHROPIC_BASE_URL=https://api.veltrixplatform.com`. Everything accepts a custom base_url. The Gateway optimizes and governs each call for you; tokens are still billed to your provider API account. Cursor IDE can't be intercepted (proprietary backend) — but Cursor CLI can.
What's the difference between Free · Lens, Otimizador and Gateway?+
It's an additive ladder — each step includes the previous one. Free · Lens is free and only gives you visibility: how much AI you spend and consume, via a Chrome extension, across 8 providers. Otimizador (R$ 39/mo) adds the Playground and effectiveness analysis — the toolkit for YOU to test prompts, compare models side by side and cut verbose tokens, stretching your subscription. Gateway (R$ 179/mo + 1.5% of spend under management) adds the control plane: it optimizes and governs calls automatically, for companies whose software or agents call the API.
Do I need to change my code?+
Only on the Gateway, and it's almost nothing: you stay on the official SDK (OpenAI Python, Anthropic SDK, etc) and swap the base_url for ours. Request, response, streaming, function calling — all the same, one line. Free · Lens and Otimizador don't touch your code: Lens is a browser extension and Otimizador is a Playground where you test prompts.
How does the 1.5% of spend under management work on the Gateway?+
You pay a floor of R$ 179/mo plus 1.5% on the spend routed through Veltrix — the 'spend under management'. That 1.5% applies only to the volume you direct through our control plane, not to your entire account at the providers. The panel shows the open, auditable calculation. And importantly: this amount covers the Veltrix service only — the tokens consumed at OpenAI/Anthropic/etc are still billed to YOUR own API account with them.
Does Otimizador compress my prompt automatically?+
No — and we make a point of being honest about it. Otimizador is a toolkit that TEACHES you to optimize: in the Playground you run the prompt, see tokens/cost/latency, compare models and spot where to cut verbose context. You're the one who applies the cut. AUTOMATIC compression, cache and routing on every call belong to the Gateway, not to Otimizador.
How much latency does it add?+
On the Gateway, p95 overhead < 50ms (compression + cache lookup). Since an LLM call normally takes 800-3000ms, this is imperceptible. On a cache hit, the response comes back in ~100ms total (no need to call the LLM). Free · Lens and Otimizador don't sit in the path of your production calls.
Are my prompts stored?+
By default we store the full prompt and response text (truncated at 10,000 characters) together with metadata (tokens, cost, model, latency) — that's what powers the Audit & Request Logs screen. That text is automatically wiped after 7 days (even within your plan's retention window); only the metadata remains after that. If you'd rather not retain the text at all, there's an opt-in toggle in Account Settings ('Data retention: metadata-only') that switches to metadata-only logging from the moment you turn it on — the Logs screen shows a placeholder instead of the text for calls made with the toggle on. Your real provider key is AES-256 and never appears in logs. Compression runs on an open-weight model (Microsoft's LLMLingua-2) hosted by us — nothing goes to a third party.
Which providers and models are supported?+
OpenAI, Anthropic, Google Gemini, DeepSeek, Mistral, Groq and Cohere — all via the OpenAI standard. Over 30 models today. New models are added as the providers release them.
Can I try it without a card?+
Yes. Free · Lens is free forever, no card. Otimizador has a 7-day free trial, also no card. On the Gateway, the first pilots run with a trial (our Gemini 2.5 Flash key) so you can measure before plugging in your own key. You only add a card when you decide to subscribe.
Can I cancel anytime?+
Yes. Self-service in the panel, any time. No penalty, no lock-in. You keep access until the end of the current month; we export your data as CSV and delete it within 30 days.
What if I have more than 15 people / mandatory SSO / an MSA contract?+
Then it's Enterprise — reach us at vendas@veltrixplatform.com. Tailored: Google/Microsoft SSO, audit logs, MSA, invoicing, volume pricing, optional on-premise.
7 days to see the savings. No card to get started.
A three-step ladder — Free · Lens (free, visibility only), Otimizador (R$ 39/mo) and Gateway (R$ 179/mo + 1.5% of spend under management). Each step includes the previous one. You choose what fits your scenario.