CLI Reference¶

The delentia command-line tool is the primary interface to the Delentia OS Control Plane SDK. (Corrected 2026-09-24: this page previously called the command rct throughout - the real entry point registered in pyproject.toml is delentia = "rct_control_plane.cli:main", confirmed against pyproject.toml and python -m rct_control_plane.cli --help's real output. rct is not an installed command.)

Install the CLI:

pip install -e .
delentia version
delentia start --ui-test
delentia init
delentia doctor
delentia --help

Global Options¶

Flag	Description
`--output json`	Output as JSON (default: table)
`--output table`	Output as rich table
`--output tree`	Tree view (for graph commands)
`--verbose`	Enable debug logging
`--version`	Print version and exit
`--help`	Show help and exit

Commands¶

`delentia version`¶

Print SDK version, Python version, and component status.

delentia version

Output (real, verified 2026-09-24 against python -m rct_control_plane.cli version):

                  Delentia OS — Version Info
┌─────────────┬──────────────────────────────────────────────┐
│ Field       │ Value                                        │
├─────────────┼──────────────────────────────────────────────┤
│ version     │ 2.2.6                                        │
│ package     │ delentia-os                                  │
│ name        │ delentia-os                                  │
│ description │ Constitutional AI Operating System SDK       │
│ python      │ 3.13.14                                      │
│ license     │ Apache-2.0                                   │
│ homepage    │ https://delentia.com                         │
│ repository  │ https://github.com/delentia-labs/delentia-os │
└─────────────┴──────────────────────────────────────────────┘

Recommended first-run order for a fresh install:

delentia version
delentia start --ui-test
delentia init
delentia doctor
delentia start

`delentia serve`¶

Start the Control Plane REST API server.

delentia serve [--port PORT] [--reload] [--workers N]

Option	Default	Description
`--port`	`8000`	Port to listen on
`--reload`	off	Auto-reload on code changes (dev mode)
`--workers`	`1`	Number of Uvicorn worker processes

Example:

delentia serve --port 8000 --reload
# → http://localhost:8000
# → Docs: http://localhost:8000/docs

`delentia compile`¶

Compile a natural language intent into an intent record.

delentia compile TEXT [--user-id USER_ID] [--context JSON]

Option	Default	Description
`TEXT`	required	The intent text to compile
`--user-id`	`"anonymous"`	User identifier for audit trail
`--context`	`"{}"`	JSON context blob

Example:

delentia compile "Refactor authentication module" --user-id alice

Output (table):

┌────────────┬──────────────────────────────────┐
│ Field      │ Value                            │
├────────────┼──────────────────────────────────┤
│ Intent ID  │ a1b2c3d4-...                     │
│ Status     │ compiled                         │
│ FDIA Score │ 0.8734                           │
│ Tier       │ S4                               │
└────────────┴──────────────────────────────────┘

`delentia build`¶

Build a DSL execution graph from a compiled intent.

delentia build --intent-id ID [--dsl-file FILE] [--output FILE]

Option	Default	Description
`--intent-id`	required	Intent ID from `delentia compile`
`--dsl-file`	auto	Path to a `.dsl` file (optional)
`--output`	stdout	Write graph JSON to file

Example:

delentia compile "Deploy microservice to staging" --user-id devops | delentia build --intent-id -
delentia build --intent-id a1b2c3d4 --output graph.json

`delentia evaluate`¶

Evaluate a compiled + built intent through policy and governance checks.

delentia evaluate --intent-id ID [--architect-score FLOAT]

Option	Default	Description
`--intent-id`	required	Intent ID to evaluate
`--architect-score`	`1.0`	Override Architect gate (0.0–1.0)

A=0 blocks output

Passing --architect-score 0.0 will cause FDIA = 0 and block all output. This is the constitutional safety guarantee — not a bug.

Example:

delentia evaluate --intent-id a1b2c3d4 --architect-score 0.9

`delentia status`¶

Show current state of a compiled intent.

delentia status INTENT_ID

Example:

delentia status a1b2c3d4-5678-...

Output:

Intent   : a1b2c3d4-5678-...
State    : COMPLETED
Created  : 2026-04-21T10:00:00+00:00
FDIA     : 0.8734
Policy   : COMPLIANT

`delentia list`¶

List recent intents.

delentia list [--limit N] [--status STATUS] [--user-id USER_ID]

Option	Default	Description
`--limit`	`10`	Max intents to show
`--status`	all	Filter by state (e.g., `COMPLETED`, `FAILED`)
`--user-id`	all	Filter by user

`delentia audit`¶

Display full audit trail for an intent.

delentia audit INTENT_ID [--output json]

Example:

delentia audit a1b2c3d4 --output json

Outputs the complete signed audit chain including FDIA score, SignedAI consensus tier, and policy decisions.

`delentia metrics`¶

Display runtime metrics for the Control Plane.

delentia metrics [--output json]

Sample output:

Metric                   Value
────────────────────────────────────
Intents processed        1,247
FDIA avg score           0.8734
Warm recall hits         89.3%
Consensus tier S4 calls  847
Policy violations        2
Uptime                   14d 06h

`delentia reset`¶

Reset in-memory state (development/testing only).

delentia reset [--force]

Destructive Action

delentia reset clears all in-memory intent state. Use --force to skip confirmation. Persisted audit logs are not deleted.

`delentia workflow run`¶

Run a DAG workflow defined in a .yaml file through the real ALGO-20 WorkflowEngine (real networkx DAG scheduling, real dependency-respecting sequential/parallel/hybrid execution) - added 2026-09-25 (Round 44 item I.1). This command builds WorkflowEngine directly rather than through the full 41-algorithm kernel, so it starts immediately (no torch/FAISS import cost).

delentia workflow run path/to/workflow.yaml [--poll-interval SECONDS]

Schema (maps 1:1 onto the engine's existing TaskConfig/WorkflowDefinition - no new semantics invented):

name: demo-workflow
mode: sequential   # sequential | parallel | hybrid
tasks:
  - id: fuse1
    type: fusion    # the only task type with a real executor today - see below
    depends_on: []
    config:
      fusion_config:
        modalities: {text: [0.1, 0.2, 0.3]}
        fusion_strategy: hybrid
  - id: fuse2
    type: fusion
    depends_on: [fuse1]
    config:
      fusion_config:
        modalities: {text: [0.7, 0.8, 0.9]}

Only type: fusion has a real executor today

WorkflowEngine._run_task_type() has a real executor for exactly one task type ("fusion" - real ALGO-19 Data Fusion). Any other type value is refused at load time by this command rather than silently accepted and run as a no-op that the engine would otherwise still mark COMPLETED (with an honest "simulated": true flag in its raw output, but the scheduler does not distinguish it from real completion when resolving downstream dependencies). Widen the allowlist in rct_control_plane/workflow_yaml_loader.py only once a real executor exists for a new type.

The command exits 0 only if every task in the workflow reaches a real completed status; it exits 1 on a YAML/validation error or if any task fails.

Removed 2026-09-24: delentia intent submit and delentia health do not exist

Both sections used to appear here but describe commands that are not registered by the real CLI - confirmed against python -m rct_control_plane.cli --help's real output (Error: No such command 'intent'/'health'). delentia status (with no argument) and delentia doctor cover overlapping ground for now; if a real equivalent to either removed command is wanted, that is new feature work, not a documentation fix.

Shell Completion¶

Enable tab-completion for bash, zsh, or fish (Click derives the completion env var name from the real program name, delentia, not rct):

# bash
eval "$(_DELENTIA_COMPLETE=bash_source delentia)"

# zsh
eval "$(_DELENTIA_COMPLETE=zsh_source delentia)"

# fish
eval (env _DELENTIA_COMPLETE=fish_source delentia)

Environment Variables¶

Removed 2026-09-24: this table was fabricated

None of RCT_PORT, RCT_LOG_LEVEL, RCT_STATE_DIR, or RCT_ARCHITECT_SCORE are read anywhere in rct_control_plane/cli.py (confirmed by grepping the real source for os.environ/os.getenv - the only real env var touched there is DELENTIA_DAEMON_ENABLED, which the start command sets, not reads as user configuration). Every option shown elsewhere on this page (--port, --reload, etc.) is a real CLI flag, not an environment variable - use those instead.

CLI Reference¶

Global Options¶

Commands¶

delentia version¶

delentia serve¶

delentia compile¶

delentia build¶

delentia evaluate¶

delentia status¶

delentia list¶

delentia audit¶

delentia metrics¶

delentia reset¶

delentia workflow run¶

Shell Completion¶

Environment Variables¶

`delentia version`¶

`delentia serve`¶

`delentia compile`¶

`delentia build`¶

`delentia evaluate`¶

`delentia status`¶

`delentia list`¶

`delentia audit`¶

`delentia metrics`¶

`delentia reset`¶

`delentia workflow run`¶