> ## Documentation Index > Fetch the complete documentation index at: https://qodex.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Projects > Projects isolate one target app's data, members, environments, memory, API keys, and BYOK settings. # Projects Projects are the boundary around one target app in Qodex. A project owns its chats, scenarios, scripts, findings, environments, memory, API keys, members, and BYOK credentials. Keeping these objects project-scoped prevents data and access from crossing between apps or teams. ## What belongs to a project Every meaningful object in Qodex is project-scoped: chats, scenarios, scripts, findings, test runs, environments, collections, memory, API keys, and BYOK credentials. Cross-project access is not possible from the agent or the API. The data model lives in Postgres. The `projects` table is the root. `project_members` joins users to projects with a role. `project_api_keys` mints scoped keys for webhooks and CI. `environments` holds target URLs, auth profiles, and constraints for staging, production, preview, or any other environment. ## Workspaces Today, workspaces are an in-memory `OrgContext` singleton that brackets a set of projects. The full org-as-persisted-tenant model is partial: organization persistence is a runtime bridge, not a database table. Per-project isolation is fully shipped; org-level rollups are evolving. ## Members and roles Users sign up with email and password and get verified by email. A user joins a project through `project_members` with a role. Membership scopes what the user can read and do in the project. The project switcher in the UI lists every project the user belongs to. ## Project-scoped routes The web app uses `/p/{slug}/...` for everything project-scoped: ```text theme={null} /p/acme/chat /p/acme/scenarios /p/acme/test-runs /p/acme/findings /p/acme/playground /p/acme/settings ``` The API mirrors this. The CLI flips between projects with `qodex projects use `. ## Environments A project can have many environments. Each environment carries: * A name and a base URL (and `apiBaseUrl` if different). * Auth: static `auth_token`, or `api_login_config` (HTTP login + JSONPath token), or `ui_login_steps` (Playwright drives login, captures storageState). * Constraints: read-only, max requests per second, allow destructive tests, allow security testing. The same scenario runs against any of them by passing `${baseUrl}`, `${apiBaseUrl}`, `${authEmail}`, `${authPassword}` and the SCREAMING\_SNAKE variants. Switching environments is one click. ## BYOK Bring-your-own-key credentials are stored at the project level. Each project can pick its provider (OpenAI today, Anthropic and Google on the roadmap) and supply its own key. Every LLM call uses the project's key directly. Qodex takes zero margin on AI spend. Platform-funded mode runs on Qodex's keys, enforced by per-project 5-hour and weekly caps. BYOK and codex\_pool traffic is exempt from those caps. ## Cost tracking Every LLM call writes one row to `llm_usage_log` with the project id, provider, model, input and output tokens, latency, outcome, and cost. The admin console at `/admin#llm` shows per-project and per-source breakdowns. You see exactly where spend is going. ## When to use it * Use one project per target app. A monolith with one frontend and one API is one project. * Create a separate project per environment only when the apps differ enough, such as different auth or domains. Otherwise use one project with multiple environments. * Create a separate project per customer if you run a multi-tenant agency setup. ## When not to use it * Splitting a single app into projects per team. Use chats and tags inside one project instead. * Sharing scenarios across projects. Today scenarios are project-bound. To share, export and re-import. ## On the roadmap Planned: organization persistence. Today `OrgContext` is an in-memory bridge. The plan is a persisted org-as-tenant layer with org-level rollups, SSO, and billing. See [TESTING-PLATFORM-ARCHITECTURE.md](https://github.com/flinket/qodeclaw/blob/master/TESTING-PLATFORM-ARCHITECTURE.md). Planned: per-project provider selection in the UI. Anthropic Claude and Google Gemini join OpenAI as supported providers. Per-project BYOK for the non-OpenAI providers is the remaining column. ## Related Project-scoped knowledge. Project-scoped targets and auth. Who can do what in a project. Per-project key management.