dibbla workflows

Build, version, and execute Dibbla workflows from the CLI — plus the supporting `nodes`, `edges`, `inputs`, `tools`, `revisions`, and `functions` sub-trees.

A Dibbla workflow is a directed graph of nodes (agents, functions, conditionals) wired together by edges. Each workflow has a definition, a set of inputs, optional API endpoints, and a revision history.

The CLI exposes the entire workflow surface as seven top-level command groups, all sharing the same global formatting flags:

Global flags

These apply to every command in this section:

dibbla workflows

list

List every workflow in the org.

$dibbla workflows list
NAME          LABEL             NODES  HAS_API
expense-flow  Expense Reporter  12     true
support-bot   Support Triage    8      false

get

Print the workflow definition. Use --revision <id> to fetch a specific snapshot.

$dibbla wf get expense-flow -o yaml > expense-flow.yaml
$dibbla wf get expense-flow --revision a71z

create

Create a workflow from a YAML or JSON definition file.

$dibbla wf create -f expense-flow.yaml
# stderr: confirmation line
Created workflow expense-flow (revision: HEAD)
# stdout: created workflow definition (yaml/json/table per --output)

update

Replace an existing workflow’s definition.

This does not create a revision. update, and every nodes / edges / inputs / tools mutation, overwrite HEAD in place — the response literally reports revision: HEAD. Only dibbla revisions create inserts a snapshot. If you want to be able to go back, take one first.

delete

Delete a workflow and all its revisions.

validate

Validate a definition without saving it. Useful in CI before update.

$dibbla wf validate -f expense-flow.yaml
# Prints the validation report — valid: true/false, node_count, edge_count
 
# In CI, test the payload. An invalid definition still exits 0:
$dibbla wf validate -f expense-flow.yaml -o json | jq -e '.valid'

execute

Execute a workflow, optionally targeting a specific API node.

$dibbla wf execute expense-flow --data '{"amount": 42}'

$dibbla wf execute expense-flow -F input.json --node api-submit

url

Print the URL of the workflow editor.

api-docs

Print readable API endpoint documentation, including HTTP method, URL, request and response schemas, and a curl example.

dibbla nodes

add

Add a node to a workflow. The node definition can come from a file or be passed inline as JSON.

One of --file or --inline is required.

$dibbla nodes add expense-flow -f nodes/classifier.yaml

$dibbla nodes add expense-flow \
  --inline '{"id":"sum","type":"function","function":"math.sum"}'

remove

Remove a node by ID.

dibbla edges

Edges connect a source node’s output port to a target node’s input port. The CLI specifies an edge as "<src.port -> tgt.port>".

add

$dibbla edges add expense-flow "classifier.out -> sum.in"

remove

Same shape as add, removes instead.

list

$dibbla edges list expense-flow
EDGE
classifier.out -> sum.in
sum.out -> approver.amount
approver.approved -> notify.message

dibbla inputs

set

Set a single input value on a node.

$dibbla inputs set expense-flow classifier model gpt-4o-mini
$dibbla inputs set expense-flow classifier system_prompt "" --null

dibbla tools

Attach or detach tools on agent nodes.

add

remove

Same arguments as add, removes instead.

$dibbla tools add expense-flow classifier web_search
$dibbla tools remove expense-flow classifier web_search

dibbla revisions

Workflow snapshots — and they are explicit only. update, nodes, edges, inputs and tools all write straight to HEAD and leave no history behind them; the editor’s autosave does the same. dibbla revisions create is the only thing that takes a snapshot.

The practical consequence: if you want to be able to undo a change, snapshot before you make it. There is nothing to recover afterwards.

Revision ids are four base-36 characters (a71z), not a prefixed hash.

list

$dibbla rev list expense-flow
ID            TIMESTAMP             LABEL
k3f9  2026-04-28T14:02:11Z  Snapshot before tool swap
a71z  2026-04-22T11:18:38Z  Initial create

create

Create a snapshot of the current HEAD.

restore

Restore a previous revision to HEAD.

Nothing snapshots the current HEAD for you. The restore overwrites HEAD in place, and whatever was there is gone — there is no roll-forward. Take a revision first:

dibbla revisions create expense-flow      # snapshot what you are about to replace
dibbla revisions restore expense-flow <revision_id>

$dibbla rev restore expense-flow a71z
Restored revision a71z for expense-flow

dibbla functions

Browse the function registry — the catalog of pre-built functions that workflow nodes can call.

list

$dibbla fn list --tag math
NAME          SERVER  DESCRIPTION            TOOLS
math.sum      core    Sum a list of numbers
math.average  core    Arithmetic mean

get

Print the full definition of a function — schema, inputs, outputs, side effects.

$dibbla fn get core math.sum -o yaml

See also

• dibbla run — local task pipelines, separate from the workflow runtime.
• dibbla deploy — deploy the app that hosts your workflow’s API endpoints.