dlin

dbt model lineage CLI. Parses SQL files directly or reads a compiled manifest.json. No Python required.

Works for developers navigating live SQL files, analysts exploring a shared manifest, AI agents via CLI prompt or MCP server, and CI pipelines.

Column-level lineage (dlin column upstream / dlin column downstream) requires a compiled manifest.json and is usable in practice while remaining Experimental for coverage reasons.

Motivation

When I edited dbt models in VS Code, dbt Power User was my go-to companion for navigating lineage. AI agents have no such companion. I watched them grep through dbt projects to find model dependencies. It works, but they end up calling grep repeatedly and relying on fragile string matching to piece together ref() and source() relationships.

dlin is designed to fill that gap: a CLI tool that lets AI agents understand a dbt project's structure without falling back to grep. It is equally useful for humans, and its stdin/stdout interface makes it easy to combine with jq, git diff, and other CLI tools.

To replace grep, speed and size matter. dlin is a small, self-contained binary with no runtime dependencies. It parses SQL directly, evaluates common Jinja patterns without Python, parallelizes file I/O, and caches aggressively.

The key idea behind dlin is that finding the right models fast is what matters most. The hard part for agents is knowing which models to look at in the first place. dlin focuses on making model-level lineage fast and provides column-level lineage for deeper analysis.

Install

Cargo (Rust)

cargo install dlin

pip / uv (Python)

For convenience, dlin is also available as a Python package. The installed binary is native and does not require Python at runtime.

pip install dlin-cli   # or: uv tool install dlin-cli

GitHub Releases

Pre-built binaries for Linux, macOS, and Windows are available on the Releases page. You can also use the installer scripts:

macOS / Linux:

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/eitsupi/dlin/releases/latest/download/dlin-installer.sh | sh

Windows (PowerShell):

powershell -ExecutionPolicy Bypass -c "irm https://github.com/eitsupi/dlin/releases/latest/download/dlin-installer.ps1 | iex"

Quick start

# Full lineage graph
dlin graph -p path/to/dbt/project

# Downstream impact analysis
dlin impact orders

# List models as JSON
dlin list -o json --json-fields unique_id,file_path

# Pipe changed files into lineage
git diff --name-only main | dlin graph -o json

Source modes

dlin supports two source modes for model-level commands.

SQL parse mode (default) is for developers editing a live dbt project. It reads ref() and source() directly from SQL without waiting for dbt compile, so model-level lineage quickly identifies the related models. Read the relevant SQL for exact transformation behavior.

Manifest mode (--source manifest) is for analysts or agents analyzing compiled state rather than editing live SQL. A developer compiles once and shares manifest.json; users explore the resolved graph without SQL or Python. Use column-level lineage when a field-level trace is needed; it requires the compiled manifest.

# SQL parse mode: reads SQL files directly (default)
dlin graph orders

# Manifest mode: reads manifest.json only
dlin graph orders --source manifest
dlin summary --source manifest

For model-name inputs, manifest.json is the only file needed in manifest mode. File-path inputs (e.g. models/foo.sql) fall back to standard dbt directory layout when dbt_project.yml is absent, which may not match projects with custom path configuration. (check-manifest always requires a full project.)

Limitations of SQL parse mode

• var() resolves from vars.yml or dbt_project.yml (--vars CLI overrides not supported)
• Runtime context (target.type, env_var()) is not evaluated
• Conditional Jinja branches use default values; non-default paths may be missed
• Generic test IDs are dlin-specific (e.g. test.not_null.orders.order_id) and do not match dbt's naming; use manifest mode when exact test IDs matter

When these limitations matter, use --source manifest.

AI agent integration

CLI approach

Recommended for developers with access to a live dbt project. Works with SQL parse mode (no dbt compile needed) as well as manifest mode.

Install dlin and add the following to your AGENTS.md, CLAUDE.md, or system prompt:

## dbt project structure analysis

Use `dlin` to explore dbt model dependencies.
Do NOT grep/cat/find through SQL files.

```bash
dlin summary                                           # Project overview (start here)
dlin graph <model> -u 2 -d 1 -q                        # Upstream/downstream lineage
dlin impact <model>                                    # Downstream impact with severity
dlin list -o json --json-fields unique_id,sql_content  # Read SQL content
git diff --name-only main | dlin graph -q              # Lineage of changed files
```

For full option reference: `dlin --help`, `dlin graph --help`, etc.

The key line is "Do NOT grep/cat/find through SQL files". Without it, agents default to familiar tools. dlin --help is designed for tool discovery, so the prompt can stay minimal.

MCP server (experimental)

For analysts and agents who work from a distributed manifest.json without access to the full project. Runs in manifest mode only.

dlin mcp exposes a stdio MCP server that AI assistants supporting MCP can connect to directly.

dlin mcp --dialect bigquery path/to/manifest.json

Available MCP tools: project summary, model search, lineage, impact analysis, and column-level lineage.

Pass --dialect to match your project's SQL dialect for accurate column lineage. Requires a compiled manifest.json (dbt compile).

Features

• SQL parse mode: single binary, no Python, no manifest.json needed for model-level lineage
• Manifest mode: works from manifest.json alone; useful for analysts or agents without a full project checkout
• MCP server (experimental): dlin mcp serves lineage data via stdio MCP for direct AI assistant integration
• Recursive upstream / downstream: -u N / -d N to control traversal depth
• Impact analysis with severity: dlin impact scores downstream nodes and flags exposure reachability
• Composable: stdin accepts model names or file paths; pipe with jq, dlin list, git diff, etc.
• Agent-friendly: --error-format json emits structured {"level","what","why","hint"} on stderr; --help is designed for tool discovery
• Column-level lineage (experimental): traces columns across models with transformation classification through the Rust lineage backend; requires manifest.json

Mermaid diagrams

dlin outputs Mermaid flowcharts that render natively on GitHub, GitLab, Notion, and other Markdown environments.

Simplified graphs with --collapse

Automatically remove intermediate nodes to see just the endpoints (nodes with no predecessors or no successors); everything in between becomes transitive "(via N)" edges:

# Collapse intermediate models; only endpoints remain
dlin graph --collapse -o mermaid

# Focal mode: keep only sources, exposures, and specified focus models
# (ignores BFS window pseudo-endpoints; works best with -u/-d limits)
dlin graph orders --collapse=focal -u 3 -o mermaid

flowchart LR
    exposure_weekly_report>"weekly_report"]
    model_combined_orders["combined_orders"]
    model_order_summary["order_summary"]
    source_raw_customers(["raw.customers"])
    source_raw_orders(["raw.orders"])
    source_raw_payments(["raw.payments"])

    source_raw_customers ==>|"exposure (via 2)"| exposure_weekly_report
    source_raw_orders ==>|"exposure (via 3)"| exposure_weekly_report
    source_raw_orders -.->|"source (via 1)"| model_combined_orders
    source_raw_orders -.->|"source (via 1)"| model_order_summary
    source_raw_payments ==>|"exposure (via 3)"| exposure_weekly_report
    source_raw_payments -.->|"source (via 1)"| model_order_summary

    classDef model fill:#4A90D9,stroke:#333,color:#fff
    classDef source fill:#27AE60,stroke:#333,color:#fff
    classDef exposure fill:#E74C3C,stroke:#333,color:#fff
    class exposure_weekly_report exposure
    class model_combined_orders model
    class model_order_summary model
    class source_raw_customers source
    class source_raw_orders source
    class source_raw_payments source

Positional focus models are always preserved during collapse, so dlin graph orders --collapse keeps orders even if it would otherwise be intermediate.

Pipe to build focused diagrams

Combine dlin list, jq, and dlin graph to extract exactly the nodes you want:

# Staging models → 1 hop downstream, models only, grouped by directory
dlin list -s 'path:models/staging' -o json | jq -r '.[].label' |
  dlin graph -d 1 --node-type model --group-by directory -o mermaid

flowchart LR
    subgraph models_marts["models/marts"]
        model_combined_orders["combined_orders"]
        model_customers["customers"]
        model_order_summary["order_summary"]
        model_orders["orders"]
    end
    subgraph models_staging["models/staging"]
        model_stg_customers["stg_customers"]
        model_stg_online_orders["stg_online_orders"]
        model_stg_orders["stg_orders"]
        model_stg_payments["stg_payments"]
        model_stg_retail_orders["stg_retail_orders"]
    end

    model_orders -->|ref| model_customers
    model_stg_customers -->|ref| model_customers
    model_stg_online_orders -->|ref| model_combined_orders
    model_stg_orders -->|ref| model_order_summary
    model_stg_orders -->|ref| model_orders
    model_stg_payments -->|ref| model_order_summary
    model_stg_payments -->|ref| model_orders
    model_stg_retail_orders -->|ref| model_combined_orders

    classDef model fill:#4A90D9,stroke:#333,color:#fff
    class model_combined_orders model
    class model_customers model
    class model_order_summary model
    class model_orders model
    class model_stg_customers model
    class model_stg_online_orders model
    class model_stg_orders model
    class model_stg_payments model
    class model_stg_retail_orders model

Column names in nodes with --show-columns

Add --show-columns to include column names inside Mermaid node labels, useful for understanding what each model produces at a glance:

dlin graph orders -u 1 -d 0 --show-columns --node-type model,source -o mermaid

flowchart LR
    model_orders["orders<br/>---<br/>order_id, customer_id, order_date, status, total_amount, payment_method"]
    model_stg_orders["stg_orders<br/>---<br/>order_id, customer_id, order_date, status"]
    model_stg_payments["stg_payments<br/>---<br/>payment_id, order_id, amount, payment_method"]

    model_stg_orders -->|ref| model_orders
    model_stg_payments -->|ref| model_orders

    classDef model fill:#4A90D9,stroke:#333,color:#fff
    class model_orders model
    class model_stg_orders model
    class model_stg_payments model

Combines well with --collapse to show rich detail on fewer endpoint nodes.

Other graph options

dlin graph orders -u 2 -d 1                            # focus on specific model
dlin graph -o mermaid --collapse --show-columns        # columns in collapsed nodes
dlin graph orders --collapse=focal -u 3 -o mermaid    # focal: sources + exposures + orders
dlin graph -o mermaid --group-by directory             # group by directory
dlin graph -o mermaid --direction tb                   # top-to-bottom layout
dlin graph --node-type source,exposure                 # filter by node type
dlin graph -o dot | dot -Tsvg > out.svg                # Graphviz rendering

Output formats: ASCII (default), JSON, Mermaid, Graphviz DOT, Plain, SVG, HTML.

Column-level lineage (Experimental)

dlin column upstream and dlin column downstream trace columns across models. They require a compiled manifest.json, so run dbt compile first. Regression fixtures and real dbt project validation show that column-level lineage is usable in practice. Experimental reflects the Rust stack's shorter production history and remaining cross-dialect and complex-SQL coverage gaps, not a prototype-only status.

# Where does each output column of orders come from?
dlin column upstream orders

# What downstream columns are affected if stg_orders.order_id changes?
dlin column downstream stg_orders --column order_id

# Mermaid flowchart
dlin column upstream customers -o mermaid
dlin column downstream stg_orders --column order_id -o mermaid

# Specific columns only
dlin column upstream orders --column order_id --column status

# Verify manifest freshness before querying
dlin check-manifest && dlin column upstream orders

Column upstream

Traces each output column of a model back to its raw source columns, following references across intermediate models.

dlin column upstream customers -o mermaid

flowchart LR
  subgraph sg0["customers"]
    n0_0["customer_id"]
    n0_1["email"]
    n0_2["first_name"]
    n0_3["last_name"]
    n0_4["lifetime_value"]
    n0_5["order_count"]
  end
  subgraph sg1["orders"]
    n1_0["order_id"]
    n1_1["total_amount"]
  end
  subgraph sg2["raw.customers"]
    n2_0["email"]
    n2_1["first_name"]
    n2_2["id"]
    n2_3["last_name"]
  end
  subgraph sg3["raw.orders"]
    n3_0["id"]
  end
  subgraph sg4["raw.payments"]
    n4_0["amount"]
  end
  subgraph sg5["stg_customers"]
    n5_0["customer_id"]
    n5_1["email"]
    n5_2["first_name"]
    n5_3["last_name"]
  end
  subgraph sg6["stg_orders"]
    n6_0["order_id"]
  end
  subgraph sg7["stg_payments"]
    n7_0["amount"]
  end

  n2_2 -->|"direct"|n5_0
  n5_0 -->|"direct"|n0_0
  n2_0 -->|"direct"|n5_1
  n5_1 -->|"direct"|n0_1
  n2_1 -->|"direct"|n5_2
  n5_2 -->|"direct"|n0_2
  n2_3 -->|"direct"|n5_3
  n5_3 -->|"direct"|n0_3
  n4_0 -->|"direct"|n7_0
  n7_0 -->|"direct"|n1_1
  n1_1 -->|"aggregation"|n0_4
  n3_0 -->|"direct"|n6_0
  n6_0 -->|"direct"|n1_0
  n1_0 -->|"aggregation"|n0_5

customer_id, email, etc. pass through stg_customers unchanged from raw.customers (all direct). lifetime_value and order_count are aggregated at the customers model. The final edge to customers is labeled aggregation, while all upstream hops carry their actual transformation type (here direct, since staging and mart models pass columns through unchanged).

Transformation types shown on edges: direct, aggregation, expression, cast, conditional, unknown.

Column downstream

Traces a column forward to all downstream models and columns that depend on it.

dlin column downstream stg_orders --column order_id -o mermaid

flowchart LR
  subgraph sg0["customers"]
    n0_0["order_count"]
  end
  subgraph sg1["order_enriched"]
    n1_0["order_id"]
  end
  subgraph sg2["orders"]
    n2_0["order_id"]
  end
  subgraph sg3["stg_orders"]
    n3_0["order_id"]
  end

  n2_0 -->|"aggregation"|n0_0
  n3_0 -->|"direct"|n1_0
  n3_0 -->|"direct"|n2_0

stg_orders.order_id flows directly into orders.order_id and order_enriched.order_id. orders.order_id is then aggregated into customers.order_count. Each edge shows its per-hop transformation type.

Performance benchmark

Correctness-gated means from the synthetic scalability run (3 runs, 1 warmup):

These synthetic results use artifacts derived from the real dbt fixture. Native Rust avoids Python startup and dlin showed lower latency here; dlin uses --no-cache, Parrant includes parsing, and dbt-meta query excludes build. They are not a general ranking; see the full results and reproduction.

Known limitations

• Requires dbt compile: no SQL parse mode fallback; manifest with compiled SQL is always needed
• SELECT * chains: resolution depends on YAML column definitions in upstream models; unresolved columns are reported in errors[]
• Dialect-specific syntax: pass --dialect bigquery (or other dialect) for better coverage
• Performance: first run parses all upstream models; results are cached in .dlin_cache/ for subsequent queries

Key subcommands

list

dlin list                                                   # all models and sources
dlin list orders -o json --json-fields unique_id,file_path  # specific model as JSON
dlin list --node-type source                                # sources only

impact

$ dlin impact orders
Impact Analysis: orders
==================================================
Overall Severity: CRITICAL

Summary:
  Affected models:    1
  Affected tests:     1
  Affected exposures: 1

Impacted Nodes:
  [critical] weekly_report (exposure, distance: 1)
  [high    ] customers (model, distance: 1) [models/marts/customers.sql]
  [low     ] assert_orders_positive_amount (test, distance: 1)

Filtering

dlin graph -s tag:finance,path:marts  # selector expressions (union)
dlin graph --node-type model,source   # filter by node type

Credits

Hard fork of dbt-lineage-viewer by Simon Muller (MIT license). The original focused on TUI-based exploration; dlin removes the TUI and targets non-interactive use: scripting, CI, and AI agents.

License

MIT