Centry deployment for Alita

Requirements

• Docker
• Docker compose (docker-compose-plugin preferred)

Configuration

1. Copy envs/default.env to envs/override.env

2. Edit envs/override.env:

   • Set APP_HOST to IP/DNS of platform
   • Set LICENSE_USERNAME and LICENSE_PASSWORD to your github username and token in order for pylons to clone private repositories automatically
   • Change passwords/keys (for production, use default ones for dev at your own risk)

   Special command for SECRETS_MASTER_KEY:

   docker run --rm --entrypoint= getcarrier/pylon:1.2.22 python -c 'from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())'

3. SSL (optional): copy your server.crt, server.key, ca.crt to ssl/, then:

   • edit pylon_*/pylon.yml: uncomment SSL-related settings
   • edit docker-compose.yml: uncommend SSL-related settings
   • edit envs/override.env: set APP_PROTO to https

4. OpenID Connect SSO (optional):

   • edit pylon_auth/configs/auth_core.yml: set auth_provider to oidc
   • edit pylon_auth/configs/auth_oidc.yml: insert config for your OIDC provider
   • edit pylon_auth/configs/auth_init.yml: specify ID from your provider (e.g. email) for default platform admin

Start

docker compose pull
docker compose up -d
docker compose logs -f

First-time platform configuration

1. Login with admin account and wait for private project to be created (~5-10 minutes, refresh the page periodically untill Public and Private projects are present in dropdown)

2. [PGVector] In Public project Settings -> AI Configuration create new PgVector credential:

   • Label: elitea-pgvector
   • Shared: NO (do not check)
   • Set as default: NO (do not check)
   • Connection string: postgresql+psycopg://POSTGRES_USER:POSTGRES_PASSWORD@pgvector:POSTGRES_PORT/POSTGRES_DB use values from your default.env/override.env; example: postgresql+psycopg://centry:changeme@pgvector:5432/db

3. [PGVector] In Administration mode Tasks /~/administration/~/tasks/tasks/ start:

   • Task: sync_pgvector_credentials
   • Parameter (input below task dropdown): force_recreate,save_connstr_to_secrets

4. [LiteLLM] In Administration mode Tasks /~/administration/~/tasks/tasks/ start:

   • Task: create_database
   • Parameter (input below task dropdown): litellm

5. [LiteLLM] In Tasks table open logs ("file" action) for created task (usually last one with meta: {'task': 'create_database'}) and copy connection string from log.

6. [LiteLLM] In Administration mode Runtime -> Remote /~/administration/~/runtime/remote/ open runtime_engine_litellm config ("file" action)

   • Click Load (Raw)
   • Change database_url from null to copied connection string
   • Generate master key in format sk-[alphanumeric] (example: sk-abcd1234ABCD)
   • Change litellm_master_key to generated master key
   • Click Save

   Example:

   database_url: postgresql://aaa:bbb@ccc:5432/litellm
   litellm_master_key: sk-abcd1234ABCD
   

7. [LiteLLM] In Administration mode Runtime -> Pylons /~/administration/~/runtime/pylons/ open pylon-predicts settings ("cog / wheel" action)

   • Click Restart
   • Wait for pylon-predicts to restart

8. [LiteLLM] In Administration mode Tasks /~/administration/~/tasks/tasks/ start:

   • Task: sync_llm_entities
   • Parameter: (leave empty)

9. [Agent state] In Administration mode Tasks /~/administration/~/tasks/tasks/ start:

   • Task: create_database
   • Parameter (input below task dropdown): agentstate

10. [Agent state] In Tasks table open logs ("file" action) for created task (usually last one with meta: {'task': 'create_database'}) and copy connection string from log.

11. [Agent state] In Administration mode Runtime -> Remote /~/administration/~/runtime/remote/ open indexer_worker config ("file" action)

    • Click Load (Raw)
    • Uncomment agent_memory_config block
    • Change connection_string to copied connection string
    • Click Save

    Example:

    agent_memory_config:
      type: postgres
      connection_string: postgresql://aaa:bbb@ccc:5432/agentstate
      connection_kwargs:
        autocommit: true