Configuration¶

Navegador has minimal required configuration. The only thing you typically need to set is where the graph database lives.

Database path¶

SQLite (default)¶

By default navegador writes a navegador.db file in the current working directory. Override with the --db flag or the NAVEGADOR_DB environment variable:

# flag (takes precedence)
navegador ingest ./repo --db ~/.navegador/myproject.db

# environment variable
export NAVEGADOR_DB=~/.navegador/myproject.db
navegador ingest ./repo

The .navegador/ directory convention keeps the database alongside your project:

my-project/
  .navegador/
    navegador.db    ← graph database
  src/
  ...

Add .navegador/ to .gitignore — the database is a build artifact, not source.

Redis (production)¶

For team or CI environments, point NAVEGADOR_DB at a Redis instance running FalkorDB:

export NAVEGADOR_DB=redis://localhost:6379

With authentication:

export NAVEGADOR_DB=redis://:[email protected]:6379

Install the Redis extra if you haven't already:

pip install "navegador[redis]"

SQLite vs Redis: when to use which¶

Both backends implement the same GraphStore interface. You can migrate by re-ingesting against the new backend.

GitHub token¶

Required for navegador wiki ingest --repo owner/repo to access private wikis or to avoid rate limits on public repos.

export GITHUB_TOKEN=ghp_...
navegador wiki ingest --repo myorg/myrepo

For public repos, wiki ingestion works without a token but will hit GitHub's unauthenticated rate limit (60 req/hr).

Project-local config¶

Drop a .navegador/config.toml in your project root for project-specific defaults:

[database]
path = ".navegador/navegador.db"

[ingest]
exclude = ["node_modules", "dist", ".venv", "migrations"]
incremental = true        # use content hashing by default
redact = false            # strip secrets from ingested content

[mcp]
read_only = false         # set true to prevent agents from writing to the graph
max_query_complexity = 100  # Cypher query complexity limit

LLM provider config¶

Configure LLM providers used by navegador ask, navegador docs, and navegador semantic-search. Requires pip install "navegador[llm]".

[llm]
provider = "anthropic"    # "anthropic", "openai", or "ollama"
model = "claude-3-5-haiku-20241022"

[llm.anthropic]
api_key_env = "ANTHROPIC_API_KEY"   # env var name (not the key itself)

[llm.openai]
api_key_env = "OPENAI_API_KEY"
model = "gpt-4o-mini"

[llm.ollama]
base_url = "http://localhost:11434"
model = "llama3"

Cluster config¶

For team deployments using a shared Redis graph with pub/sub, task queue, and session coordination:

[cluster]
enabled = true
redis_url = "redis://redis.internal:6379"
graph_name = "navegador-team"

[cluster.pubsub]
channel = "navegador:events"

[cluster.queue]
name = "navegador:tasks"

[cluster.sessions]
ttl_seconds = 3600

See the Cluster mode guide for full setup instructions.

Environment variable reference¶