Identify the caller
A Gateway Key maps to one Project. Provider credentials never reach the application.
SELF-HOSTED LLM GATEWAY
Give every token an owner.
Halro sits between your applications and model providers. Applications keep a Gateway Key and a public model alias; Halro controls provider credentials, budgets, routing, redaction, audit, and usage accounting.
REQUEST BOUNDARY
Policy runs before Provider I/O.
Each client Request and Provider Attempt remains separate, so retries and failures do not disappear into one ambiguous record.
Enforce policy
Validate the Route, model, CIDR, field rules, rate limits, concurrency, and budget before Provider I/O.
Select a target
A Route binds versioned Deployments and Provider Profiles behind a stable model alias.
Settle with evidence
Record tokens, cost, latency, pricing evidence, retries, and the final outcome for every request.
BOUNDARIES
A focused gateway, not an agent platform.
Use Halro for: centralized provider credentials, stable routes, budgets, request policy, cost attribution, and audit evidence.
Do not treat it as: agent orchestration, workflow execution, memory, an evaluation platform, or a replacement for provider quota controls.
Current deployment model: a standalone single-writer instance. Experimental capabilities are labelled next to the feature.
NEXT STEP