Free forever, one provider · 10-day trial on any paid plan
Simple pricing, aligned with your AI maturity.
Start free on one provider — real-time proxy and attribution included, no credit card, no expiry. Move to Base to observe everything, Growth to enforce and optimize, Business for full CFO-grade accountability. Upgrade only when you see the value.
Try it
Free
See real attribution on one provider — no card, no expiry.
Start for freeWhat's included
- Real-time proxy with attribution
- 1 provider connection · 1 seat · 1 API key
- Spend dashboard (feature & model)
- 1 budget — Alert, Degrade, or Block
- Emergency Freeze — one-click kill switch
- Image generation (DALL-E, Imagen)
- 7-day data retention
- Velocity circuit breaker
- Email alerts for budgets and anomalies
- Environments: production vs staging vs CI spend
- $50/mo tracked spend — proxy keeps working past that
Observe
Base
Observe and monitor your AI spend, unlimited providers.
Start my 10-day trialEverything in Free, plus
- Unlimited provider connections — incl. AWS Bedrock, Vertex AI, custom endpoints
- Role-based access control — Owner, Admin, Developer, Finance, Viewer
- Department breakdown · CSV export
- Up to 3 budgets, any enforcement mode
- Waste detection
- Chargeback & GL mapping
- AI ROI Summary & Budget vs Actuals reports
- Compromise Risk signals
- CI cost gate: fail builds on cost regressions
- MCP server: ask about spend from Claude Code or Cursor
- Slack alerts · no 7-day retention limit
Enforce + Optimise
Growth
For teams actively reducing AI waste and enforcing budgets.
Start my 10-day trialEverything in Base, plus
- Unlimited budgets
- One-click recommendation apply
- Smart routing rules · Failure Loop Breaker
- Auto Router — complexity-based model routing
- Load balancing across models and providers
- Experiments: compare prompt versions on cost
- Saved views & custom charts
- Langfuse & OpenTelemetry export
- Multi-key balancing across API keys
- OpenAI Batch API with cost attribution
- Live billing pull for OpenAI and Anthropic
- Prometheus metrics endpoint
- Session & user-level attribution
- Cost per business outcome
- FOCUS-aligned export
- Scheduled report delivery by email
- Investment vs. Waste classification
- Token Maxing Detector · Context Tax Analyser
- 30/60/90-day spend forecast
- FinOps Maturity Score
- Semantic + exact response caching
- Prompt cache + batch routing recommendations
- Proactive nightly insights
- Executive View
- Shadow Spend Reconciliation
Govern + Advise
Business
Full CFO-grade accountability and AI advisory.
Start my 10-day trialEverything in Growth, plus
- AI Cost Advisor — natural language queries
- "Why did spend increase 40%?" answered
- Agentic Cost Simulator
- What-If Scenario Planner
- MCP / A2A workflow attribution
- AI Efficiency Score — cost per outcome, tracked monthly
- Single sign-on (SAML) — Okta, Entra ID, Google Workspace
- Require SSO for every member, with verified domains
- Audit log with search and CSV export
- SCIM user provisioning (e.g. Okta, Entra ID)
- JWT authentication from your identity provider
- Provider keys in AWS, Vault or Azure Key Vault
- MCP gateway with tool policy
Free never expires. Every paid plan includes a 10-day free trial with full Business-plan access. Start free →
Compare every feature
Every row below is a real, shipped capability — not a roadmap item. Nothing here reflects a feature that doesn't exist yet.
Core platform
Free
Base
Growth
Business
Real-time proxy with attribution
Feature, department, user, session, GL account, tier — always fully on, every plan
OpenAI, Anthropic, Google, Mistral, Groq, Together AI, Azure OpenAI, AWS Bedrock, Google Vertex AI, and any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio) — 1 connected at a time on Free, unlimited from Base up
Invite teammates into your account — 1 on Free (solo); 3 on Base; 10 on Growth; 25 on Business
14-day rolling baseline — spend, call-frequency, and rejection-rate spikes, with root-cause hypothesis
Budget-threshold and anomaly/security alerts to up to 10 addresses, with per-address unsubscribe
Budget, anomaly, security and report alerts via an incoming webhook
1 on Free; up to 3 on Base; unlimited on Growth and above
Degrade / Block budget enforcement
Every plan, including Free — Free is capped at 1 budget, not enforcement mode
Tag calls with X-Cost-Environment (production, staging, ci, eval...), see spend per environment and set a budget on any one of them
Fail a CI build when a change makes your test run more expensive than the baseline; one HTTP call from any CI
'Key' scope + parent-budget chaining — org → department → user → key ceilings, tightest wins
Data retention
7 days on Free; no retention limit on paid plans
Retry, mismatch, and context-bloat detection
Pattern-based PII/secrets redaction and prompt-injection screening, incl. tool-output scanning
AI ROI Summary & Budget vs Actuals reports
On-demand PDF / interactive view
One-click kill switch for all proxied spend — every plan, including Free
Audio, rerank and Responses API endpoints
/v1/audio/transcriptions, /speech, /v1/rerank (Cohere, Jina, Voyage) and /v1/responses, with the same budgets, guardrails and cost tracking as chat
Image generation
DALL-E, gpt-image-1, Imagen — priced per image, full budget/velocity coverage
Attribution & guardrails
Free
Base
Growth
Business
Rewrite a feature's model to a cheaper one, always or only when the prompt is under a token count
Auto-retry on a different connected provider on timeout, 429, or 5xx
Stops a session after 3 identical tool calls in a row or a run of consecutive errors; alert-only or stop
Token- and dollar-denominated — fixed 10x baseline auto-trip on either, or set manual TPM/$-per-min caps
Independent proxy rate limit
~1 req/sec on Free, to keep it from serving production traffic
MCP / A2A workflow attribution
Per-tool, per-agent, per-handoff cost breakdown
New-source-IP / new-model-family detection per proxy key
Cost intelligence & optimization
Free
Base
Growth
Business
One model name; each request classified simple / medium / complex / reasoning and routed to the tier model you chose — explained per request, savings measured against your top model
Tag calls with X-Cost-Variant and compare prompt versions, models or environments on cost per call, tokens, latency, errors and spend per outcome. Not a quality eval
Crawl / Walk / Run, calculated in real time
Model growth and see which budgets would breach
1–10x call multiplier, growth %, 30/90-day projection
Reporting & governance
Free
Base
Growth
Business
Email the AI ROI Summary PDF monthly (previous month), weekly or on the last day of the month (month to date), with optional approval before it goes out
Conformance validation in progress
CFO-level dashboard with unit economics
Build charts from any metric, split and filter, share them with the team and pin them to the Overview
MCP server for your spend data
Ask Claude Code, Cursor or any MCP client about spend, budgets and CI results. Read-only
Natural-language Q&A over your own spend data
Cost per outcome in dollars, tracked month over month
Month-to-date spend, requests, tokens, latency and budget utilisation, scrapeable into Prometheus / Grafana
One model name spread across models or providers by weight, round-robin or lowest cost, skipping failing deployments
Langfuse & OpenTelemetry export
Send every call, with cost, tokens, latency and attribution tags, to Langfuse or any OTLP backend
Several API keys per provider, balanced by weight and rate-limit aware, so one key's limit stops being your ceiling
Create batches through Cognocient; the discounted cost is attributed to your tags when each batch finishes
Compares provider-reported billing against proxy-observed spend. Daily live pull for OpenAI and Anthropic with an admin key; manual or CSV entry for other providers
Security & access
Free
Base
Growth
Business
Owner, Admin, Developer, Finance and Viewer roles enforced on every API call; admin-only by default for anything new; proxy keys cannot manage team, billing, SSO or provider keys
Okta, Microsoft Entra ID, Google Workspace or any SAML identity provider; DNS-verified domains; optional require-SSO for all members (the owner keeps break-glass access)
Who changed what and when — team and role changes, provider keys, budgets, routing, SSO, exports, and denied attempts — searchable, with CSV export. Request contents are never stored
Okta, Microsoft Entra ID or another SCIM 2.0 identity provider creates, updates and removes team members automatically; deprovisioning ends access immediately
Authenticate with short-lived JWTs from your own identity provider instead of long-lived keys; claims become cost attribution
Provider keys in your own vault
AWS Secrets Manager, HashiCorp Vault or Azure Key Vault: keys fetched at request time, never stored by Cognocient
Govern MCP servers with tool allow/deny policy, argument scanning, per-call pricing, budgets and attribution
Frequently asked questions
Is the Free plan actually free forever?
Yes — no trial clock, no credit card, no expiry. It's capped at 1 provider connection, 1 budget (any enforcement mode — Alert, Degrade, or Block), 7-day data retention, and $50/mo of tracked spend, but the real-time proxy and attribution never turn off, even past that cap — your app keeps working.
What notifications do I get on the Free plan?
Every plan, including Free, gets email alerts for budget thresholds and anomalies, plus the dashboard and Emergency Freeze. Slack alerts start on Base. The Free plan is single-seat, so there is no team to invite.
Which plans include SCIM, JWT authentication, secret-manager keys and the MCP gateway?
Business. Multi-key balancing, the OpenAI Batch API and the live billing pull for Shadow Spend are on Growth and above. Audio, rerank and the Responses API work on every plan, like embeddings and image generation.
What happens after my tracked spend passes $50/mo on Free?
Your app keeps working exactly as before — the proxy never stops forwarding calls to your AI provider. Cognocient just stops writing new attribution records for the rest of the billing cycle. Upgrade any time to keep tracking, or wait for the next cycle.
Can I try a paid plan before paying?
Yes. Any paid plan (Base, Growth, or Business) offers a 10-day free trial with full Business-plan access, no credit card required. Start it from the Free plan any time — your data and configuration carry over either way.
Can I upgrade or downgrade later?
Yes, from Settings → Billing, which opens the Paddle billing portal. Your plan in Cognocient changes as soon as Paddle confirms the change.
What happens if I exceed 3 budgets on the Base plan?
You'll be prompted to upgrade to Growth for unlimited budgets. Existing budgets and their enforcement keep working — you just can't create a 4th until you upgrade.
Does routing calls through Cognocient add latency?
Our published benchmark (/docs/latency-benchmark) measured roughly 2-8ms added at the median with 1-10 simultaneous requests to one instance. At 50-100 simultaneous requests to a single instance it grows to hundreds of milliseconds, because the backend runs as a single worker today. If latency matters more than pre-call enforcement, the Python wrapper reports usage after each call and adds none.
How are my AI provider API keys secured?
Provider keys are encrypted at rest and are never returned by the API or shown in full in the dashboard. Your app only ever holds a Cognocient proxy key, which can be revoked independently at any time. On Business, keys can stay in your own AWS, HashiCorp or Azure vault instead.
Is there a setup fee or long-term contract?
No setup fee and no annual lock-in — all paid plans are billed monthly. Integration is a base_url change in your existing OpenAI SDK (or any OpenAI-compatible client), not a migration project.
Which plans include single sign-on, roles, and the audit log?
Role-based access control (Owner, Admin, Developer, Finance, Viewer) is included on every plan that has team seats — Base and up. Single sign-on (SAML 2.0 with Okta, Microsoft Entra ID, Google Workspace or any SAML identity provider, verified domains, and an optional require-SSO policy) and the searchable, exportable audit log are on the Business plan.
Can I route to models I host myself, like Ollama or vLLM?
Yes. Add a Custom endpoint for any OpenAI-compatible server (Ollama, vLLM, LM Studio, llama.cpp, a LiteLLM gateway) and send model=custom/<name>. It must be a public https:// URL, because Cognocient calls it from the cloud. Self-hosted calls are tracked at $0 unless you enter your own per-token rates. AWS Bedrock and Google Vertex AI are supported natively too.
What is the Auto Router?
Send one model name and Cognocient classifies each request as simple, medium, complex or reasoning, then routes it to the model you assigned to that tier — with a per-request explanation and measured savings against your most expensive model. It runs inside the proxy with no extra LLM call. Available on Growth and above.
Can I put a hard cap on my AI spend, even on the Free plan?
Yes — Emergency Freeze is a one-click kill switch on every plan, including Free, that blocks all proxied calls immediately regardless of budgets. It can also auto-freeze on a spend-velocity spike or a high-severity compromise signal if you opt in.
More questions? See the refund policy or read the quickstart docs.
Get started today
Try every feature free for 10 days.
Full Business-plan access, no credit card, no sales call. When the trial ends you drop to the Free plan on one provider, unless you choose a paid plan.