Solutions
Everything the platform does.
Build on every major model, understand every dollar, and scale without a rewrite. This page walks the whole system, in depth.
Just want to integrate? See how it works.
Build
One key and one endpoint puts every major provider behind the client you already run.
Understand
Every request is attributed, so spend reads by team, project, and model instead of by invoice.
Scale
Routing, orchestration, and reserved capacity keep throughput up when demand and providers wobble.
02 / The reality
Every team built its own AI stack.
Marketing, support, engineering, finance: each with their own models, clouds, keys, and bills. It all works. And nobody can say what ran, where, or what it cost.
Hicap Platform
Now they all run on one.
Hicap replaces this with one endpoint, one invoice, one set of rules.
One unified API layerLower costs
Higher reliability
Nothing unaccounted
One key. Total control.
One key. Every major provider.
Point your existing OpenAI-compatible client at api.hicap.ai and the whole catalog is live. Change a model string to change providers; nothing else in your code moves.
1curl https://api.hicap.ai/v1/chat/completions \2 -H "api-key: $HICAP_API_KEY" \3 -H "Content-Type: application/json" \4 -d '{5 "model": "gpt-5.4",6 "messages": [7 { "role": "user", "content": "Hello" }8 ]9 }'
One key
A single credential opens every major provider. No per-provider contracts, quotas, or key rotation calendars.
Drop-in compatible
The OpenAI-compatible clients you already run work unchanged: chat, streaming, tool calls, and embeddings.
One bill
Usage across every provider lands on a single invoice, priced and itemized the same way everywhere.
Know where every dollar went.
Tag every request to a feature, team, app, or customer, and drill from the org down to the individual API key. Finance gets chargeback without the spreadsheet; platform owners get budgets that enforce themselves.
Model Usage
Cost, volume & tokens by model across all apps
Top Models by Credits
Spend leaderboard
#1 eleven_multilingual_v2
29,161.00
#2 eleven_v3
27,708.00
#3 scribe_v1
3,046.00
#4 scribe_v2
2,658.00
#5 claude-opus-4.7
1,191.00
Usage attribution
Every call tagged to a feature, team, app, or customer.
Per-call logging
Request, response, model, tokens, and latency on every call.
Quota enforcement
Hard usage limits per org, app, user, and key.
Policy enforcement
Rate limits, model allow-lists, and routing rules at the gateway.
Scoped keys
Connection-scoped, rotatable keys with per-key limits.
Full audit trail
Every request and admin action retained and exportable.
The same work, off the frontier bill.
Visibility is only half the story. Because Hicap sits on the request path, it does not just report the bill, it changes it: routine work moves to cheaper models and the ledger shows the price each request avoided.
Routing ledger
Cost / 1M tokens
classify
-$7.30
gpt-5.5 → llama-4-405b
$9.40 → $2.10
summarize
-$8.30
claude-opus-4-8 → claude-sonnet-4-6
$12.60 → $4.30
extract
-$4.90
gemini-3.1-pro → mistral-large-3
$7.80 → $2.90
reason
kept
claude-opus-4-8 → claude-opus-4-8
$12.60 → $12.60
billed $21.90 of $42.40 list
saved $20.50
Route down by default
Routine work runs on models that cost a fraction of the frontier and clear the same quality bar.
Frontier where it counts
Hard reasoning stays on the strongest models. The policy decides per request, not per project.
Savings you can audit
Every routing decision is logged with the price it avoided, so the number on this ledger is checkable.
Smarter routing means better performance for less.
Every request is automatically routed to the best model for its needs, whether that's lower cost, faster speed, or higher quality. Your top models are only used when they're truly needed.
incoming request
↑
One policy shapes all your traffic.
Routing picks the best model for each request. Orchestration works a level above: set the frontier and open-source mix, chain smaller models to do the job of one big one. No rewrite, no new keys, no migration.
prefer: "balanced"
p50 390ms
Your app
Hicap
Closed · frontier
Open source
Same endpoint, same key, same logs. Change the policy and the next request follows it: swap a frontier model for an open one without touching application code.
8 strategies: cost · latency · model-match · blue/green · priority + fallback · auto-failover on 429 / 5xx / timeout, with per-provider health tracking.
Reserved throughput, priced below pay-as-you-go.
Pool reserved GPU capacity for provisioned performance at up to 25% below pay-as-you-go pricing, with on-demand bursting and automatic fallback.
Select your cloud
Choose the models and providers that fit your workload, and reserve the throughput you need.
Route automatically
Requests draw from reserved capacity first. On-demand catches the burst when demand spikes.
Resell what's idle
Unused reservation lists on the spot market and the revenue flows back to you automatically.
Built to stay up when a provider goes down.
Scale is not just more tokens per second. It is throughput that holds while providers rate-limit, degrade, and recover underneath you.
Automatic failover
Requests retry across providers on 429s, 5xx errors, and timeouts. Your app sees an answer, not an incident.
Provider health tracking
Latency and error rates are watched per provider, and traffic drains away from a degrading one before it fails.
Reserved-first throughput
Reserved capacity absorbs your baseline at a fixed price. On-demand catches the burst above it.
One endpoint through it all
Failover, bursting, and provider swaps happen behind api.hicap.ai. Application code never learns about the weather.
The one-line migration
Bring order to your AI stack in one line of code.
Change a base URL and every model, every provider, and every dollar runs through one place.