Navegador

Your codebase + everything your team knows about it — in one queryable graph.

Navegador parses your source code into a property graph and layers your team's knowledge on top: decisions, concepts, rules, people, wiki pages, and meeting outputs. AI coding agents get structured, precise context instead of raw file dumps.

navegador — Spanish for navigator / sailor

Two layers, one graph

┌─────────────────────────────────────────────────────────────────┐
│  KNOWLEDGE LAYER                                                │
│  Concepts · Rules · Decisions · WikiPages · People · Domains    │
│                                                                 │
│         ↕  GOVERNS / IMPLEMENTS / DOCUMENTS / ANNOTATES         │
│                                                                 │
│  CODE LAYER                                                     │
│  Repository · File · Module · Class · Function · Method         │
│  Variable · Import · Decorator · (call graphs, hierarchies)     │
└─────────────────────────────────────────────────────────────────┘
              stored in FalkorDB  (embedded local · Redis prod)

The code layer is built automatically by ingesting source trees. The knowledge layer is populated by your team — manually, via wiki ingestion, or from PlanOpticon meeting analysis output.

Quick start

pip install navegador

# Ingest your repo
navegador ingest ./myrepo

# Load context for a file
navegador context src/auth.py

# Search across code + knowledge
navegador search "rate limit" --all

# Explain a symbol
navegador explain AuthService

# Check graph stats
navegador stats

MCP integration

Add to your Claude / Cursor / Gemini MCP config:

{
  "mcpServers": {
    "navegador": {
      "command": "navegador",
      "args": ["mcp", "--db", ".navegador/graph.db"]
    }
  }
}

Available MCP tools:

For a federated workspace, aggregate repo graphs and serve the super-graph:

navegador aggregate backend=/path/a frontend=/path/b --db central.db
navegador mcp --db central.db --read-only
# or roll up at startup:
navegador mcp --db central.db --federate backend=/path/a --federate frontend=/path/b

Query/context/search tools accept an optional repo argument to scope to one namespace; omit it to span all repos (e.g. cross-repo blast radius).

Knowledge layer

Beyond code structure, navegador stores what your team knows:

# Record an architectural decision
navegador add decision "Use FalkorDB for graph storage" \
  --rationale "Cypher queries, SQLite-backed zero-infra mode"

# Define a business concept and link it to code
navegador add concept PaymentProcessing
navegador annotate PaymentProcessing --function process_charge

# Add a rule
navegador add rule "All writes must go through the service layer"

# Ingest your GitHub wiki
navegador wiki ingest --repo myorg/myrepo

# Import PlanOpticon meeting analysis
navegador planopticon ingest ./meeting-output/

Graph schema

Code nodes: Repository · File · Module · Class · Function · Method · Variable · Import · Decorator

Knowledge nodes: Concept · Rule · Decision · Person · Domain · WikiPage

Edges: CONTAINS · DEFINES · IMPORTS · CALLS · INHERITS · REFERENCES · DEPENDS_ON · GOVERNS · IMPLEMENTS · DOCUMENTS · ANNOTATES

Storage

from navegador.graph import GraphStore

store = GraphStore.sqlite(".navegador/graph.db")    # embedded
store = GraphStore.redis("redis://localhost:6379")  # shared server

Navegador resolves the backend from layered configuration — --db/--redis-url, then NAVEGADOR_REDIS_URL/NAVEGADOR_DB, then the project's .navegador/config.toml, then ~/.config/navegador/config.toml, then the embedded default. navegador doctor prints which layer decided and whether the result actually works.

Note: .navegador/graph.db is a FalkorDB (Redis) RDB snapshot, not a SQLite database — sqlite3 cannot open it. The sqlite name refers to the embedded, single-file mode provided by falkordblite. Inspect the graph with navegador (e.g. navegador query), not with SQLite tooling.

A shared server, without Docker

Once you work across several repos — or run more than one agent — a single resident graph beats every process re-reading its own copy from disk:

navegador server install     # official FalkorDB module + service, no Docker
navegador server status      # version, memory, resident graphs

install also writes ~/.config/navegador/config.toml, so every project uses the shared server unless it configures otherwise. Existing local graphs move across without re-ingesting, verified on both sides:

navegador scan ~/repos                            # what exists, and what is stranded
navegador storage migrate --all --root ~/repos    # copy it all into the server

Offline documentation

The full documentation ships inside the package — no network, no mkdocs:

navegador manual                       # list every page
navegador manual quickstart            # read one
navegador manual --search "redis"      # search all of them

Agents get the same pages through the read_docs MCP tool.

Language support

Framework enrichment

After ingesting code, navegador can promote generic AST nodes to framework-specific semantic types:

navegador enrich                          # auto-detect frameworks
navegador enrich --framework django       # target a specific framework

Supported frameworks: Django, FastAPI, React / Next.js, Express.js, React Native, Rails, Spring Boot, Laravel

Structural analysis

navegador impact AuthService --depth 3    # blast radius
navegador trace handle_request            # execution flow from entry point
navegador deadcode                        # unreachable functions/classes
navegador cycles                          # circular dependencies
navegador testmap                         # link tests to production code
navegador diff                            # map uncommitted changes to graph
navegador churn .                         # behavioural coupling from git history

Intelligence layer

navegador semantic-search "authentication flow"   # embedding-based search
navegador communities                              # detect code communities
navegador ask "what calls the payment service?"    # natural language queries
navegador docs src/auth.py                         # generate documentation

Requires an LLM provider: pip install navegador[llm]

Python SDK

from navegador import Navegador

nav = Navegador.sqlite(".navegador/graph.db")
nav.ingest("./myrepo")
nav.add_concept("Payment", description="Payment processing", domain="billing")

results = nav.search("auth")
bundle = nav.explain("AuthService")
owners = nav.find_owners("AuthService")

Cluster mode (agent swarms)

For multi-agent setups sharing a Redis-backed graph:

navegador init --redis redis://host:6379 --cluster

Features: shared graph with local snapshots, pub/sub notifications, task queues, distributed locking, session namespacing, checkpoints, agent messaging, observability dashboard.

Additional integrations

navegador codeowners ./myrepo             # parse CODEOWNERS → ownership graph
navegador adr ingest docs/decisions/      # Architecture Decision Records
navegador api ingest openapi.yaml         # OpenAPI / GraphQL schemas
navegador deps ingest package.json        # external dependency tracking
navegador pm ingest --github org/repo     # GitHub issues → knowledge graph
navegador editor setup claude-code        # generate MCP config for editors
navegador explore                         # browser-based graph visualization

Installation

PyPI

pip install navegador

Standalone binaries

No Python required — download prebuilt binaries from GitHub Releases:

Navegador is POSIX-only — its embedded graph backend (falkordblite) has no native Windows build. On Windows, run navegador inside WSL2 and use the Linux binary; CI runs the full test suite under WSL2 on every change to packaging. Use Ubuntu 24.04 or newer, since navegador needs Python 3.12+. See Installation for details.

From source

git clone https://github.com/ConflictHQ/navegador.git
cd navegador
pip install -e ".[dev]"
pytest

Contributing

See CONTRIBUTING.md. Bug reports and feature requests welcome via GitHub Issues.

License

MIT — CONFLICT