Guarani AI
ENFR Start Free Trial

Guarani Mesh

Private beta · Trial workspaces on request

Enterprise multi-model gateway and FinOps router

One inference endpoint for every model your teams use. Mesh routes each request to the right reasoning model, fails over across regions and providers, and shows what every team spends.

Start Free TrialTalk to an engineer

What it does

  • Unified OpenAI-compatible API

    Unified OpenAI-compatible API in front of Azure OpenAI, Amazon Bedrock, Google Gemini models on Gemini Enterprise Agent Platform and open-weight models on dedicated GPUs

  • Dynamic routing

    Dynamic routing by latency, cost, context length and data residency

  • Multi-region failover

    Multi-region failover with health checks and retries

  • Contextual caching

    Contextual caching for repeated context

  • Per-team cost tracking

    Per-team cost tracking, budgets and alerts

What routing looks like

Sample data for the Balanced routing policy: traffic share, latency and spend by provider over 24 hours, then spend by team against its daily budget.

Routing policyRouting policy: BalancedSample data
Traffic share, latency and spend by provider over 24 hours, for the selected policy
ProviderSharep50p95Spend / 24 h
Azure OpenAIeastus2, swedencentral38%218 ms540 ms$77.33
Amazon Bedrockus-east-1, us-west-231%236 ms575 ms$47.32
Gemini 4 Argon (Google Cloud)us-central1, 1M context21%208 ms520 ms$50.75
GPU (open-weight)dedicated endpoint10%172 ms390 ms$7.00
Weighted p50
217 ms
Weighted p95
532 ms
Spend, 24 h
$182.40

Spend by team, daily budget

  • payments$61.1093% of daily budget
  • search$44.2074% of daily budget
  • platform$38.7077% of daily budget
  • mobile$24.9062% of daily budget
  • data$13.5045% of daily budget

How it fits the release verdict

Every change runs as a traced pipeline across managed cloud regions, GPU inference and real device pools. Customer code is processed in isolated tenants and never used to train models.

See the whole architecture

  1. Mesh routes model calls

    Every agent call goes through Guarani Mesh. When Azure OpenAI degrades in eastus2, the router marks the call REROUTED to swedencentral and the trace continues.

Guarani Mesh

  • Router
  • Context cache
  • Budget guard
  • Telemetry

Verdicts come from versioned policy rules applied to the findings: the same findings and the same policy always produce the same verdict.

Configure it

guarani.yml declares the Mesh policy and providers.

Not publicly available yet: the CLI, SDKs and API are part of the private beta.

Read the docs: Configuration, REST API

guarani.yml, excerpt
mesh:
  policy: balanced  # latency | balanced | cost
  providers: [azure-openai, amazon-bedrock, google-gemini-enterprise, gpu-open-weight]
  residency: us     # us | eu
  daily_budget_usd:
    payments: 66
    search: 60

The Mesh endpoint follows the OpenAI chat completions format, so existing OpenAI client libraries work by changing the base URL to https://api.guarani-llc.com/v1/mesh. Use "model": "auto" to let the policy choose, or name a model your workspace allows. metadata.team attributes spend to a team budget.

POST /v1/mesh/chat/completions
curl https://api.guarani-llc.com/v1/mesh/chat/completions \
  -H "Authorization: Bearer $GUARANI_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model": "auto", "messages": [{"role": "user", "content": "Summarize this diff"}], "metadata": {"team": "payments"}}'

Data handling and deployment

Data handling

  • Encryption in transit and at rest.
  • Processing in US or EU regions chosen per workspace.
  • No training on customer code, prompts or outputs.
  • Retention you configure for repository content and artifacts.

Deployment options

  • Multi-tenant cloud with isolated tenants, the default for every plan.
  • Dedicated VPC on Azure, AWS or Google Cloud.
  • Isolated GPU inference endpoints for open-weight models.

Explore the platform