AI Gateway

gigadrive ai has two halves. models and chat run inference through the Gigadrive Network AI Gateway; usage, budgets and policies read and write an organization's spend controls.

Models

gigadrive ai models list takes no arguments and no flags.

gigadrive ai models list

anthropic/claude-haiku-4-5  Claude Haiku 4.5 (latest)  (ctx 200000)
openai/gpt-4o  GPT-4o  (ctx 128000)
openai/gpt-4o-mini  GPT-4o mini  (ctx 128000)

gigadrive ai models get <model-id> prints one model as JSON, including provider, capabilities (chat, streaming, tools, vision, json_mode), contextWindow, maxOutputTokens, routingProviders and zdrProviders.

gigadrive ai models get openai/gpt-4o

Models describes what the catalog covers.

Chat

gigadrive ai chat sends exactly one user message and prints the reply. --model is required.

gigadrive ai chat "Write a release note for a cache invalidation fix" --model openai/gpt-4o

Without --stream the command waits for the whole completion, then prints the assistant message, and prints it as JSON when the content is not a plain string. With --stream, content chunks go to stdout as they arrive and a newline closes the output.

Pass - to read the prompt from stdin:

gigadrive ai chat - --model openai/gpt-4o --stream < CHANGELOG.md

There is no flag for a system prompt, a temperature or a token limit. Use the SDK when you need one.

Usage

The three usage subcommands are scoped to an organization. --org wins; otherwise the CLI takes organizationId from .gigadrive/project.json and fails when the link carries none. All three accept --org/-o, --app/-a, --from and --to, where --from and --to are ISO 8601 timestamps and both bounds are inclusive.

summary prints JSON: a summary object with requestCount, successCount, errorCount, inputTokens, outputTokens, totalTokens, billableCostMicros and avgLatencyMs, plus byModel, byProvider and byUser breakdowns.

gigadrive ai usage summary --from 2026-07-01T00:00:00Z --to 2026-08-01T00:00:00Z

requests prints one line per gateway request, newest first, then a count of the request lines. The status is one of pending, success, error or cancelled.

2026-08-12T09:14:22.000Z  openai/gpt-4o  success  1841 tok  9205µ$
2026-08-12T09:12:07.000Z  openai/gpt-4o-mini  error  0 tok  0µ$

2 request(s).

The CLI sends no page size, so the API's default of the 100 most recent matching events applies. Use export for a range that runs longer than that.

export produces CSV and takes the maximum page size of 500 events by default. It writes to stdout unless you add --out, which takes a path and prints Wrote <path>. instead.

gigadrive ai usage export --from 2026-07-01T00:00:00Z --to 2026-08-01T00:00:00Z --out july.csv

Budgets

gigadrive ai budgets get accepts --org and prints the budget list as JSON.

gigadrive ai budgets set replaces the entire budget set for the organization. It reads JSON from --file, or from stdin when --file is absent or -, and accepts either a bare array or an object of the form { "budgets": [...] }. Every per-budget field is optional: applicationId, userId, period (day or month), maxCostMicros, maxRequests, maxTokens, maxRequestCostMicros and enabled.

cat > budgets.json <<'JSON'
[
  { "period": "month", "maxCostMicros": 50000000, "enabled": true },
  { "period": "day", "applicationId": "0195c11c-0000-7000-8000-000000000042", "maxRequests": 10000 }
]
JSON
gigadrive ai budgets set --file budgets.json

Because it is a replace and not a merge, an empty array clears every budget:

echo '[]' | gigadrive ai budgets set --org 0195c0aa-0000-7000-8000-0000000000ff

Budgets explains what each cap does when it is hit.

Policies

gigadrive ai policies get accepts --org and --app, and prints the policy as JSON or the line No policy set.

gigadrive ai policies set accepts --org and --file only. To target a single application, put applicationId in the JSON body; there is no --app on this subcommand. The body is sent as given, so a typo in a field name is not caught locally.

cat > policy.json <<'JSON'
{
  "allowedModels": ["openai/gpt-4o", "openai/gpt-4o-mini"],
  "requireZdr": true,
  "allowFallbacks": false,
  "maxOutputTokens": 4096
}
JSON
gigadrive ai policies set --file policy.json

Setting allowedModels or allowedProviders to null allows everything. Policies covers the enforcement rules.