FAQ

The getting-started FAQ covers what an account can do today and the boundaries of the public API. This page answers the broader questions about accounts, models, agents, data, billing, privacy, and self-hosting that come up after that.

Getting started

How do I get started?

Start with the Quickstart for signing in and making a first model call, use the common task index to find a specific guide, and read How Redgold works for how the system fits together. The getting-started FAQ covers what an onboarded account can use today.

How do I sign in, and which identity providers are supported?

Sign-in uses OAuth 2.0 / OpenID Connect through an external identity provider, so Redgold never sees your provider password. An interactive client establishes an account session at sign-in; treat that session as a user credential and do not reuse it as a long-lived automation secret. See authentication.

Accounts and API keys

How do I get an API key?

Create an sk-rg- key from the API-key page in the signed-in web application. The full value is shown once, so store it in a secret manager or environment variable. Minting a key requires an account session; an existing API key cannot authorize creating another key. See authentication.

My request was rejected with 401 — what is wrong?

The model API reads the credential only from the Authorization: Bearer header (and the session cookie for the web application); a key sent only as X-Api-Key is rejected. A missing, malformed, revoked, or unknown key also returns 401. An account that has run out of model-API credits returns 402 instead.

How do I limit or revoke a key?

Give each automation its own key so it can be revoked independently, and set the lowest practical per-key request limit for unattended clients. Revoke a key from the same API-key page when it is no longer needed or may have been exposed. See credential handling.

Models and the API

Which models can I call, and how do I find them?

Query /v1/models with your key and use an identifier it returns for your account; the available set changes independently of client code, so read it from that endpoint rather than hard-coding a list. redgold-flint and redgold-flint-base use /v1/chat/completions and /v1/responses; the other identifiers in your list use /v1/messages. See the Model API.

Which client libraries work?

The OpenAI SDK works against https://api.redgold.ai/v1 for Chat Completions and Responses, and the Anthropic SDK works against https://api.redgold.ai (it appends /v1/messages itself). Present the key as a bearer token in both cases; the Anthropic path reads Authorization: Bearer, not the SDK's default x-api-key. Worked examples are on the Model API page.

Is there a Redgold-specific SDK for the data platform?

The stable public client surface is the HTTP model API and the MCP server. A dedicated data-platform SDK is not published for external use yet; internal clients exist but are withheld from the public reference until their contracts and compatibility expectations are ready.

Agents and sessions

What is a coding-agent session?

A coding agent runs in an isolated workspace tied to a repository and a conversation, with a filesystem and a bounded set of tools. Session history keeps the messages and tool activity inspectable so you can review or continue the work later. An agent can delegate a bounded subtask to a child agent whose result returns to the parent conversation. See AI and agents.

Will an agent take actions without my approval?

Agent actions requested through the MCP server — spawning a session and sending a prompt — are approval-gated: the call creates an approval request attributed to your account and returns an approval id, and the action runs only after that approval resolves. Read tools are not gated. See the MCP server.

Can I reproduce an agent run exactly?

No. Session history is persisted and auditable, but model generation, external services, repository state, and a changing dataset can make a replay produce different output. A reproducible evaluation needs pinned inputs, model, configuration, code revision, and a stable data snapshot. See persistence and replay.

Data and pipelines

What is a dataset, and what is a pipeline?

A dataset is a named collection of typed records that retain source and ingestion metadata, so a caller can tell when an event occurred apart from when the platform received it. A pipeline describes how records enter the platform, how transforms run, and how a result is returned or stored. See Data and pipelines and the glossary.

Can I query my own data over the API?

The MCP server exposes read-only tenant tools that list, load, and run read-only SQL against datasets your account owns. The broad generic dataflow authoring and execution interfaces are still being finalized as external contracts, so they are not yet part of the supported public API.

Is my data visible to other accounts?

Scoping is enforced on the server from the account resolved at the edge, not from an argument in the request: MCP dataset and agent tools operate only on what your account owns, and a call naming a dataset you do not own is refused. Redgold does not sell your data or share it with third parties for their own purposes. See the MCP scoping guarantees and the Privacy Policy.

Billing and limits

How is usage measured and billed?

Usage is metered by resource consumption — agent runtime, model tokens, storage, and request counts — and used to operate the service, enforce plan limits, and bill usage-based charges. Specific numeric limits vary by plan and are shown in the account interface. Some quota controls are still being rolled out on the public path and are omitted from the docs until they are enforced. See limits.

What rate limits apply to the API?

A key may have a requests-per-minute cap enforced over a rolling 60-second window; exceeding it returns a 429 whose message states the retry interval. An upstream provider can also return 429 during a capacity event, and that status propagates to the caller without a forwarded Retry-After. Honor Retry-After when present and otherwise use bounded exponential backoff with jitter. See limits and retries.

How do I check my remaining credits?

GET /v1/credits returns the available model-API credit balance for the authenticated account, as a micros integer and a USD value. See the Model API.

Privacy and data handling

Where do my prompts go when I run a model?

Running a model sends your prompt and the context needed to answer it to the selected model provider acting as a subprocessor — Anthropic or OpenAI for external models, or Redgold's own infrastructure for self-hosted models, where the prompt is not sent to an external provider. Redgold selects providers that do not train on API data by default. See the Privacy Policy.

How do I delete my data?

Delete your account from the account settings in the app; that removes your profile, revokes the stored sign-in, and removes your conversations, content, and derived records. Audit and billing records that must be retained for legal reasons are the exception. See retention and deletion.

Self-hosting

Can I self-host Redgold?

Yes. In a self-hosted deployment, the operator controls the infrastructure, subprocessor selection, and retention, and can restrict inference to self-hosted models so no prompt leaves their infrastructure. The hosted Privacy Policy describes the managed service; a self-hosted operator publishes their own. See self-hosted deployments.

How are applications deployed?

Deployment connects a source repository to container builds, a registry, generated deployment state, and Kubernetes, with ArgoCD reconciling the generated state rather than applying direct changes to a live cluster. Tenant, registry, domain, and ArgoCD provisioning are currently completed during onboarding while public self-service controls are consolidated. See managed deployment.

MCP and integrations

How do I connect an MCP client?

Point a streamable-HTTP MCP client at https://api.redgold.ai/api/mcp and send your sk-rg- key as a bearer token. With Claude Code:

claude mcp add --transport http redgold https://api.redgold.ai/api/mcp \
  --header "Authorization: Bearer $REDGOLD_API_KEY"

The tenant MCP endpoint is a preview surface, so confirm availability for your account before wiring it into automation. See the MCP server.

What can the MCP server do?

The tenant profile exposes read-only dataset tools (list, load, schema, and read-only SQL over datasets you own) and a scoped view of your own agent sessions, tickets, and approvals. Spawning a session and prompting one are approval-gated; everything else is read-only. Operator and cluster tools are not part of this surface and are unreachable with an sk-rg- key. See the MCP server.