postgres-storage

PostgreSQL conversation storage

Persist conversations to PostgreSQL instead of the default local JSONL files. Useful when several infer processes — chat, the channels-manager daemon, the web terminal — should share one conversation store.

docker-compose.yml runs the Inference Gateway plus a PostgreSQL container; .infer/config.yaml points the CLI at both. The CLI creates its schema automatically on first connect (via infer migrate, run implicitly), so there is nothing to set up in the database by hand.

Run it

# 1. Provider keys for the gateway
cp .env.example .env   # then edit — set at least one provider key

# 2. Start the gateway + PostgreSQL
docker compose up -d

# 3. Chat — conversations are written to PostgreSQL
infer config set agent.model anthropic/claude-4.5-sonnet
infer chat

(Install the CLI first if needed: flox activate -- task install, or the install script from the repo root.)

Verify persistence

# From the CLI — lists conversations read back out of PostgreSQL
infer conversations list

# Straight from the database
docker compose exec postgres \
  psql -U infer -d infer_conversations -c \
  'SELECT id, title, count, updated_at FROM conversations ORDER BY updated_at DESC;'

Conversations survive docker compose down && docker compose up -d because the data lives in the postgres_data volume. docker compose down -v wipes it.

Configuration

The relevant block in .infer/config.yaml:

storage:
  enabled: true
  type: postgres
  postgres:
    host: localhost
    port: 5432
    database: infer_conversations
    username: infer
    password: infer
    ssl_mode: disable   # local container has no TLS; use "require" for managed PG

Any field can be overridden by an environment variable, which is the safer place for a real password:

export INFER_STORAGE_TYPE=postgres
export INFER_STORAGE_POSTGRES_PASSWORD='…'

SQLite and Cloudflare D1 use the same single-table schema and are drop-in alternatives — set storage.type to sqlite or d1 (see ../../docs/conversation-storage.md).