AI Gateway · Guardrails · LLM FinOps

One control plane for every model call your company makes

Cloptima sits in the request path. It checks policy, redacts sensitive data, reserves budget, and prices each call on an attributed ledger before the provider sees a token.

Start free trialView pricing

No credit card required • Full features • Cancel anytime

# Keep your SDK. Change the base URL.
client = OpenAI(
-   base_url="https://api.openai.com/v1",
+   base_url="https://api.cloptima.ai/v1/ai",
    default_headers={"X-Cloptima-App": "support-copilot"},
)
Request pathSample traffic
POST/chat/completionssupport-copilotclaude-sonnet-5-5

“Refund order 4471 for [EMAIL], paid with card ending [CARD].”

  1. IdentityVirtual key · team-support
  2. Policysupport-copilot + org baseline
  3. GuardrailsPII redacted · email, card
  4. BudgetReserved $0.0041 of $45.00/day
  5. CacheSemantic miss · stored after reply
  6. Routeanthropic · claude-sonnet-5-5
  7. LedgerSettled $0.0038 · team-support / prod
Gateway overhead 8.4 msAllowed · redacted
$2,806.67
Spend this month
402,500
Requests
4.2%
Blocked for safety
$695.89
Saved by caching
  • 14:02:11docs-searchCache hit
  • 14:02:09code-review-agentAllowed
  • 14:02:08sales-assistantBlocked
Bring your own keys forOpenAIAnthropicGemini and Vertex AIAmazon Bedrock250+ models in the catalog →
In the request path

Every control runs before the provider sees a token

One policy decides what a call may do. The gateway enforces it inline, so a blocked call costs nothing and a redacted one never leaks.

Guardrails

Secrets and PII leave the prompt before the prompt leaves you

Prompt · support-copilot4.1 ms

Hi, I’m [NAME]. My email is [EMAIL] and my API key is [SECRET]. Why was I charged twice?

Your own rulesPrompt injectionAzure AI Content SafetyAWS Bedrock GuardrailsGoogle Model Armor

Redact secrets and PII with your own rules, catch prompt injection, and run Azure, AWS, and Google safety scans before a request reaches a provider.

Hard budgets

Limits that block, not only alert

70%$2,806.67 of $4,000
support-copilot$45/dayBlock
code-review-agent$30/dayBlock
eval-harness$10/dayAlert

Daily and monthly limits per policy, per-request token caps, and alert-only limits for experiments.

Caching

Repeat prompts served from cache

$695.89saved this month
Provider prompt cache$526.94
Exact cache$114.31
Semantic cache, net$54.64

Exact and semantic response caches cut token spend and response time, on top of the provider's own prompt cache.

Failover

Ordered routes, priced before each try

anthropic · claude-sonnet-5-5429
bedrock · claude-sonnet-5-5931 ms

Every attempt is reserved first, so a retry never pushes spend past a hard limit.

Overhead

p95 up to 13ms

Policy, guardrails, budget, and ledger run in one pass on every call.

Attribution

Spend by team, app, environment, model, and your own fields

support-copilot$902.21
code-review-agent$625.30
docs-search$440.17
sales-assistant$360.40
invoice-extractor$214.84

Every call lands on an attributed ledger priced at your negotiated rates, so finance and engineering read the same numbers.

MCP tool governance

Decide which MCP servers and tools your agents may use

github.search_codeAllowed
postgres.execute_sqlDenied
stripe.create_refundDefinition changed

Register the servers your agents use, allow or deny each tool on every request, and review a tool's definition when it changes.

Edge gateway

Keep traffic in your VPC, under the same policy

Your VPCEdge gateway, your keys, local cache
signed policy →
← usage
Cloptima cloudPolicies, ledger, console

Run the gateway inside your network with your own keys and a local cache. It takes signed policy snapshots and reports usage back.

Models

One endpoint for every provider you hold keys for

Anthropic$1,879.87
OpenAI$543.97
Vertex AI$276.40
Amazon Bedrock$106.43

OpenAI- and Anthropic-compatible endpoints for OpenAI, Anthropic, Gemini, Vertex AI, and Amazon Bedrock, with 250+ models in the catalog.

The console

What your team sees on day one

Spend, traffic, and safety blocks fill in from the first request. There is no tagging project and no dashboard to build.

app.cloptima.ai/llm
Cloptima AI dashboard: spend, traffic, safety blocks, models, apps, budgets, and caching savings
New · Cloptima Duet

The AI teammate that follows your rules

Tag @Duet in Slack. It works on your code in its own sandbox and opens draft pull requests, and every model call runs through your org's policies and guardrails.

Meet Cloptima DuetGet early access
  • Only the repos you grant
  • Never your keys
  • Your guardrails on every call
  • Hard limits, set by your org
Built for agents

Your agents can read this site and work with the gateway

Every page serves markdown. The CLI and the MCP server use scoped tokens, so an agent can query spend, issue virtual keys, and review violations the way you would.

  • /llms.txtSite map for agents
  • api.cloptima.ai/v1/mcpMCP server, bearer token only
  • /.well-known/agent-skills/index.jsonCLI and MCP skills
bash
# Any page on this site, as markdown
curl -H "Accept: text/markdown" https://cloptima.ai/pricing

# Attribute a call with two headers
curl https://api.cloptima.ai/v1/ai/chat/completions \
  -H "Authorization: Bearer $CLOPTIMA_API_KEY" \
  -H "X-Cloptima-App: support-copilot" \
  -H "X-Cloptima-Team: team-support" \
  -d '{"model": "claude-sonnet-5-5", "messages": [...]}'

# A repeat request answers from cache
x-cloptima-cache: hit
Same ledger

Cloud, Kubernetes, and warehouse costs too

Multi-cloudAWS and GCP spend mapped to teams without manual tagging.KubernetesPod-level cost with safety-scored rightsizing.PR cost impactCost comments on GitHub pull requests before merge.BigQueryPer-query cost and alerts on scan-heavy queries.
Customers

Teams that put the gateway in front of production

−38%
“We routed our agent workflows through Cloptima’s AI Gateway with zero code changes. Dynamic fallbacks and semantic caching cut our monthly OpenAI and Anthropic bill by 38% without quality loss.”
Siddharth RaoHead of AI Engineering
1 day
“Cloptima flagged a recurring BigQuery join that was costing us thousands every week. We optimized the query in a day, and the savings paid for the platform in the first month.”
R ChandrashekharData Architect
Same day
“When one of our core services started burning through CPU, Cloptima’s anomalies caught the spike instantly. It correlated the cost jump to a deploy that morning and traced it straight to the GitHub PR.”
Ketan ChandakCTO, Adster, Inc.
<5 minfrom signup to the first governed call
p95 up to 13msfully governed request overhead
250+models in the catalog

Put your first governed call through in five minutes

Point your SDK at the gateway, bind a policy, and watch the first call land on the ledger.

Start free trialTalk to the founders
No credit card required
5-minute setup
Free trial