Skip to main content

Configuration

For an installed Scalable deployment, stage typed settings through the CLI, validate them, then apply the reviewed change:

./flow-steward config set --public-host flow.example.test --public-port 443
./flow-steward config validate
./flow-steward config apply --confirm

config set does not change the running system. A restart does not apply staged configuration. The public base URL is generated from typed host/port and TLS settings, not set as an arbitrary environment variable. Compact installations must follow their supported provider/profile instructions; the Scalable config family is not automatically available there. See CLI Reference.

For contributor development setup only, start from the environment example:

cp .env.example .env

Core installation does not require an LLM credential. After bootstrap, an administrator creates an account-scoped credential profile in the application and selects permitted provider/model combinations for projects or agents. Do not treat process-wide provider environment keys as the primary credential model, and do not assume an unavailable model silently falls back to another provider; execution reports model_unavailable.

Optional environment overlays​

The application repository includes additional examples for specific concerns:

  • .env.example.test
  • .env.example.quality
  • .env.example.ops
  • .env.example.memory
  • .env.example.prod

Keep the attachment and redaction policy block from the example files intact. Runtime defaults for those lists bootstrap from the canonical environment examples.

Contributor memory setup​

Canonical memory records live in Postgres. Optional vector and graph services are indexes. The following raw Compose example is for a contributor checkout, not the installation or configuration-update path for an installed deployment.

The standard Compose stack builds the canonical Postgres image with Apache AGE, pgvector, pg_trgm, and unaccent.

docker compose -p flow-steward -f infra/docker/docker-compose.yaml up -d --build

The candidate's canonical public deployment uses the Postgres unified profile, with pgvector and Apache AGE in the platform Postgres image. Neo4j variables and the old memory-neo4j Compose profile are not in the pinned environment manifest and are not a supported public setup path.

Built-in MCP client configuration​

Configure an MCP client using that client's supported settings and the Flow Steward MCP endpoint. The pinned public CLI has no setup-codex-mcp command; an internal repository note that names one is stale and is tracked as an application-owner follow-up.

Export a platform token before launching Codex:

export MCP_BEARER_TOKEN=<platform_pat_or_machine_token>

Use a narrowly scoped PAT or service-account token. Never disable MCP auth outside an isolated local-development environment.

Full environment reference​

See Environment Variables for a grouped reference of the main .env.example, production, ops, quality, test, provider, MCP, storage, auth, monitoring, and docs exposure variables.