Sign in to the Business Cloud dashboard, open API keys and create a key. The key is shown once — store it safely. Every key is scoped to your user and, where applicable, your organisation's budgets and model policies.
curl https://taas.cloudsigma.com/v1/chat/completions \
-H "Authorization: Bearer $TAAS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "auto", "messages": [{"role": "user", "content": "Hello!"}]}'Use model: "auto" to let the autorouter pick the best model for each request, or any model id from the model catalogue.
The signed-in Get Started wizard turns this guide into a short, safe workflow. It reads the models and policies available to your selected key, but never receives or stores the key secret.
Recipes follow the model type. The wizard offers text and agent harness recipes for compatible chat models, plus JSON or local-file examples for supported embeddings, reranking, images, speech, transcription, OCR, document conversion and speaker tasks. Select the integration key first: its model and geography policy determines the live choices. A missing recipe can mean the selected model does not support that endpoint.
Migrate and verify. The OpenClaw migration preview reads your JSON configuration locally and prints a proposed diff with rollback data. Review it before applying it. After any harness change, make a small request and check Request Log for the model, supplier, tokens and outcome. Opening a recipe or copying a command does not count as a successful request. Reopen Get Started to check verified API activity and effective payer readiness.
Business Cloud provides the following API interfaces using your TaaS API key and its model, budget and routing policies:
POST /v1/chat/completions (plus embeddings, rerank, audio and more; see the API reference). Use a compatible OpenAI SDK with the TaaS base_url.POST /v1/messages and POST /v1/messages/count_tokens. Works with the Anthropic SDKs by setting the base URL; the key is accepted via x-api-key or Authorization: Bearer.POST /v1/codex/responses. Configure Codex with base URL https://taas.cloudsigma.com/v1/codex and wire_api = "responses"; see the Codex setup. This route preserves the native Codex request format.# Anthropic format — any AI MC model, including "auto"
curl https://taas.cloudsigma.com/v1/messages \
-H "x-api-key: $TAAS_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{"model": "auto", "max_tokens": 1024, "messages": [{"role": "user", "content": "Hello"}]}'Chat Completions and Messages support streaming (SSE) and translate supported request features for the selected model. Tool use, images and other capabilities depend on the model and route. Use the native Codex endpoint for Codex harness traffic and select a permitted model from the live catalogue.
Each model and supplier has its own default caching policy. In Models, the compact Default caching row shows the first default supplier policy; expand it to see supported supplier alternatives. OpenAI and Anthropic show their corporate API defaults. Cache retention is a preference where supported, not a promise that every repeated request will hit.
For POST /v1/chat/completions, prompt caching follows the selected route's default when supported. Omit the top-level cache object, or set it to {"mode":"auto"} without a TTL or key, to keep that default. For a deliberate override, replace the model placeholder below with an id returned by /v1/models whose cache_options supports the requested mode, TTL and key.
{
"model": "MODEL_FROM_V1_MODELS",
"cache": {"mode": "explicit", "ttl": "1h", "key": "support-agent-v1"},
"messages": [{"role": "user", "content": "Hello!"}]
}cache.mode: "auto" (the default) preserves automatic caching behaviour.cache.mode: "preferred" asks for a supported preference without claiming an explicit create/extend operation.cache.mode: "explicit" requires a cache.ttl or cache.key and is intended for a deliberate create/extend operation.cache.mode: "disabled" asks to turn caching off; it is accepted only on routes that support disabling it.Supported controls vary by model route. A malformed envelope returns 400 invalid_cache_preference; an unsupported preference returns 400 cache_setting_unsupported. To restore the default, omit cache or replace it with {"mode":"auto"}, removing the TTL and key. disabled cannot include a TTL or key. A TTL can create or extend a billed cache entry, so repeat it only when you intend that operation.
Use a stable, opaque key for a reusable prompt prefix (1–256 characters without control characters). Never include credentials. Do not combine cache with provider-native controls such as prompt_cache_key, prompt_cache_options or cache_control. Reported cache reads and writes appear in usage metadata and billing. See the API reference for the accepted fields.
The catalogue search and filters stay visible as you scroll. Expand a model's supplier row to compare execution locations and the configured routing/failover order. Ask an administrator to set supplier overrides in Models & Suppliers → Suppliers / Routing order; set geographic routing preferences on the API key. Organisation policy still applies. A catalogue filter only changes what you see.
Reference price is the model creator's hosted API price (the corporate API for OpenAI and Anthropic). Best platform price identifies the supplier and comparable input/output rates available through TaaS. Expand the price table for other suppliers. Savings depend on your input/output mix, applicable tier and selected route; unavailable or incomparable prices are not shown as a discount. Non-token services retain their own billing units.
Set model: "auto" and Business Cloud classifies each request (task type, difficulty, context size, tool use) and routes it to the best candidate. Variants:
auto — balanced best fit.auto:price_perf — price/performance: genuinely cheaper models win when quality stays within a guarded fit floor.auto:cost, auto:ttft, auto:tps — cost- or latency-first selection backed by live route telemetry.auto:orchestrator / auto:subagent — role hints for agentic pipelines: orchestrators weight reasoning higher, subagents weight task fit. Combinable, e.g. auto:price_perf:subagent.Admins can enforce an algorithm at organisation, department, API-key or user level from the Autorouter page, which also shows routed traffic, selection shares and per-request routing detail.
The Quality page compares recorded fit, price/performance, blended and classifier-confidence scores by model or task. Choose a period and metric, then inspect the sample counts and comparison data. These scores help identify workloads to trial; validate results with your own acceptance tests. The timezone dropdown initially follows your browser and can be changed for consistent team reporting.
Organisations group users under shared budgets, model policies and administration. Org owners manage members and roles, per-org model overrides, budgets, and organisation integrations (identity sync). Domain verification lets an organisation claim an email domain so new sign-ups join automatically. AI MC platform admins additionally manage suppliers, system keys and all organisations.
Link your corporate IdP so users are provisioned and deprovisioned en masse, with IdP groups mapped to Business Cloud roles. Business Cloud implements standard SCIM 2.0, supported natively by Microsoft Entra ID and Okta.
Workflow (Dashboard → Org Integrations):
Microsoft Entra ID: Enterprise applications → New application → Create your own → Provisioning → Automatic; Tenant URL = the SCIM base URL, Secret token = your SCIM token; assign users/groups; start provisioning.
Okta: Applications → Browse App Catalog → SCIM 2.0 (OAuth Bearer Token) app; SCIM connector base URL = the SCIM base URL, OAuth Bearer Token = your SCIM token; enable Provisioning to App (create/update/deactivate); assign people/groups.
Role mappings: map IdP group names to Business Cloud roles (member, org admin, org owner) on the same page; membership changes in the IdP flow through on the next sync.
Bring your own supplier API key for your organisation's requests to that supplier. Your OpenAI key is used with the OpenAI corporate API. Manage keys on API keys → BYOK (organisation owners; AI MC admins can manage any organisation via the org selector). Keys are write-only — only a masked hint is shown after saving.
Select an organisation before adding or replacing a supplier key. Use the supplier choices offered in BYOK, save the credential, then verify the masked hint and a small supported request. Your application continues to authenticate with its TaaS key; the saved supplier credential is used behind that boundary. Keep enough platform funding for any permitted fallback.
Optionally set a budget per key (weekly, monthly, quarterly, annually or lifetime). Leave it blank for no cap. When this budget is exhausted, subsequent requests use standard Business Cloud billing until the period resets. Supplier charges still apply on the supplier account when its key is used.
BYOK availability depends on the supplier and endpoint. If restricted BYOK-first, platform-first or BYOK-only controls are shown, a saved preference is not proof of activation: use the displayed eligibility and registration state. These controls are not a general promise of native, streaming or media support.
Claude Code speaks the Anthropic Messages API, which Business Cloud serves natively at /v1/messages — pointing it at Business Cloud is two environment variables:
export ANTHROPIC_BASE_URL="https://taas.cloudsigma.com"
export ANTHROPIC_AUTH_TOKEN="sk-taas-..." # your AI MC API key
export ANTHROPIC_MODEL="auto" # optional: or any AI MC model id
claudeFor a guided local setup that prompts for the key, lists available models and backs up your shell file, review and run:
curl -fsSL https://taas.cloudsigma.com/static/setup/claude-code-cloudsigma.sh | bashModel choice: auto engages the Business Cloud autorouter per request; agentic runs work well with auto:orchestrator for the main loop. All usage is metered against your Business Cloud key and budgets as normal.
Codex CLI and the Desktop app-server share the same $CODEX_HOME/config.toml provider surface. The native Responses route keeps Codex-specific request semantics intact while the model selector exposes the live models allowed by your key.
/v1/models, asks for a default model, and backs up the existing Codex config.model_provider = "cloudsigma"
model = "auto"
[model_providers.cloudsigma]
base_url = "https://taas.cloudsigma.com/v1/codex"
wire_api = "responses"
supports_websockets = false
[model_providers.cloudsigma.auth]
command = "/bin/cat"
args = ["/path/to/mode-0600/taas-api-key"]For a guided local setup that prompts for the key, lists the live model library and backs up config.toml, review and run:
curl -fsSL https://taas.cloudsigma.com/static/setup/codex-cloudsigma.sh | bashThe endpoint called by this configuration is POST https://taas.cloudsigma.com/v1/codex/responses, authenticated with Authorization: Bearer. The base URL ends at /v1/codex; the client appends /responses. For configuration field meanings, see the official OpenAI configuration reference.
Keep the key outside config.toml. The setup script is a local file edit and model-list check; it does not send your Codex configuration to Mission Control.
OpenClaw can use Business Cloud through either API format. The simplest setup registers Business Cloud as an OpenAI-compatible provider:
// openclaw config — models section
{
"models": {
"providers": {
"taas": {
"baseUrl": "https://taas.cloudsigma.com/v1",
"apiKey": "sk-taas-...",
"api": "openai-completions",
"models": [{ "id": "auto", "name": "Automatic routing" }]
}
}
}
}Set the agent's model to taas/auto (or a specific model id). For Anthropic-format integration instead, use base URL https://taas.cloudsigma.com with an anthropic provider type — both routes hit the same models, budgets and billing.
Usage is metered per token at the prices on the model catalogue, including discounted cache reads where the upstream supports caching. Budgets can be set at organisation, department, API-key and user level over the standard periods — weekly, monthly, quarterly, annually or lifetime/one-off — and BYOK supplier keys accept the same per-key budgets. The dashboard shows live spend against each budget; renewal reports can be enabled per organisation.
Organisation billing: an authorised administrator selects individual or centralised billing and the central billing user in Organisation billing settings. With centralised billing, Billing names the account that pays for your requests. You do not need to fund your personal account while that policy applies. Personal funding, auto-top-up and payment controls are disabled; Minimise saves your collapsed view across sessions. Individual billing restores those controls and makes them visible again.
Admin view: organisation settings come first. The wallet funding account's balance, current auto-top-up settings and saved payment method are read-only here. Payment history can be paged through in full. Choose a concrete organisation before changing its payer or budget; All organisations is an aggregate view.
Usage starts with a shared overview of spend, requests, token volume, cache hit rate and blended cost per million tokens, including the previous period where comparable data exists. The overview stays the same across tabs. Charts initially show the top ten items; expand to inspect all available series.
Use the custom report export to group token use by model and day or month for your selected period and scope. Saved reports retain your filters and recheck access when opened or exported. Request Log provides per-request attribution; a dash for cache information means it was not established, rather than zero cache usage. Multi-lane Anthropic and OpenAI attribution uses the simplified oath, corp, AWS and Openrouter labels.
See also: API reference. This guide is updated as features ship.