Archie And AI Settings
Configure account-wide AI providers and models used by Archflow
Archflow supports Bring Your Own Key provider configuration for Archie, documentation generation, analysis, and other AI-assisted work. Open Settings → Archie and AI from the global navigation.
Self-Hosted Organizations
In a single-org installation, an administrator configures shared AI under Settings → Organization → Configure organization AI. That configuration serves the web application and background jobs; personal and project provider overrides are not used. For setup, internal model endpoints, and privacy considerations, see self-hosted configuration. The account-level provider instructions below apply to hosted accounts.
Supported Providers
| Provider | Notes |
|---|---|
| Azure OpenAI | Azure-hosted deployments and enterprise configuration |
| OpenAI | OpenAI chat and reasoning models |
| OpenRouter | Access to models from multiple providers through one API |
| Anthropic Claude | Claude model family |
| Mistral AI | Mistral-hosted models |
| HuggingFace | Hosted open-source models |
| Ollama Cloud | Ollama Cloud or a compatible Ollama endpoint |
The model list available to your account depends on the provider, credentials, endpoint, and provider-side access.
Add A Provider
- Open Settings → Archie and AI.
- Add a provider and enter its required credentials and endpoint information.
- Test the connection.
- Choose the chat model.
- Optionally choose a separate tool model for structured tool calls.
- Save the configuration.
Archflow masks stored credentials after entry. Never place provider keys in project descriptions, Architecture Documents, source uploads, screenshots, or MCP instructions.
Multiple Providers
You can keep more than one provider configuration for redundancy or different workloads. Confirm which configuration is active before running expensive generation or analysis.
Provider and model selection is currently account-wide in the user interface. Project Settings configure Archie automation, sharing, and MCP/API access; they do not expose a separate project model selector.
Archie Preferences And Memory
The same settings area includes account-wide Archie behavior and memory controls. Project-specific evidence and durable knowledge live under Documentation → Project Knowledge within each project.
Web Search With Ollama Cloud
Archie can use web search when the active AI configuration resolves to Ollama Cloud with a valid cloud key. Search uses that configuration's credentials; a local Ollama endpoint does not provide the cloud search service, and there is no separate search-provider fallback.
In managed mode, a served web search uses five managed credits. BYOK searches use no managed credits, though provider charges still apply. A message can make at most three web-search attempts. Search failures release the reserved search credits.
Use external search for outside information and project tools for the actual architecture model. A web result does not establish a fact about your project's implementation.
Usage
Open Settings → AI usage to inspect allocation, activity, usage by tool, and account limits. In managed mode, each model request uses one credit, including model failures; served web searches use five. Other tool executions and BYOK calls use no managed credits. Provider billing remains governed by the provider account associated with your credentials.