Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.knowledgestack.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

The Knowledge Stack workflow system lets you build automated ETL processes that use AI to extract, process, and output content. Each workflow is defined by four parameters — called ABCD — that control what data goes in, how it is processed, and where results are written.

​
ABCD Parameters

Every workflow is configured with these four parameters:
ParameterRoleWhat You ProvideDescription
A — SourcesInput dataFolders or documents (1-20)The content your workflow reads from
B — InstructionsProcessing rulesFolders or documents (1-20)Documents that describe how to process the data
C — OutputsDestinationFolders only (1-20)Where the workflow writes its results
D — TemplateOutput formatA single document version (optional)A template that defines the structure of the output

​
Parameter Rules

  • Each parameter’s items must be strictly disjoint — no item can appear in more than one group
  • No item can be an ancestor or descendant of another item within or across groups
  • You need read access for A, B, and D parameters, and write access for C

​
Creating a Workflow

​
Define a Workflow

POST /v1/workflow-definitions
Provide a name, description, runner configuration, and your ABCD parameters. The system validates all parameters when you save the definition.

​
Workflow Definitions API

MethodEndpointDescription
POST/v1/workflow-definitionsCreate a new workflow
GET/v1/workflow-definitionsList all workflows
GET/v1/workflow-definitions/{id}Get a specific workflow
PUT/v1/workflow-definitions/{id}Update a workflow
DELETE/v1/workflow-definitions/{id}Delete a workflow and all its runs

​
Running a Workflow

​
Invoke a Workflow

POST /v1/workflow-definitions/{id}/invoke
When you invoke a workflow:
  1. The system validates all ABCD parameters (paths exist, permissions are correct, source folders contain documents)
  2. A workflow run is created with status PENDING
  3. A scoped JWT is minted for the runner
  4. The runner is notified to begin execution
Only one run can be active (RUNNING) per workflow definition at a time. Additional invocations are queued as PENDING and dispatched automatically when the active run completes.

​
Run Status

StatusDescription
PENDINGQueued, waiting for a slot
RUNNINGCurrently executing
COMPLETEDFinished successfully
FAILEDFinished with an error

​
Workflow Runs API

MethodEndpointDescription
GET/v1/workflow-definitions/{id}/runsList runs for a workflow
GET/v1/workflow-runs/{id}Get a specific run
DELETE/v1/workflow-runs/{id}Delete a run (not while running)
POST/v1/workflow-runs/{id}/callbackRunner callback endpoint

​
Idempotency

You can include an Idempotency-Key header when invoking a workflow. If a run with the same key already exists in PENDING or RUNNING state, the API returns 200 OK with the existing run instead of creating a duplicate.

​
Self-Hosted Runners

For the current version, workflows use self-hosted runners. When you create a workflow definition, you configure a runner URL where the workflow engine sends execution requests.

​
How Runners Work

  1. Knowledge Stack POSTs the ABCD parameters and a scoped JWT to your runner URL
  2. Your runner processes the data according to the instructions
  3. When finished, your runner calls back to Knowledge Stack with the result

​
Runner Security

  • Scoped JWT: Runners receive a purpose-built token that is bound to the specific run, with a TTL matching the maximum run duration
  • SSRF prevention: Runner URLs are validated against blocked IP ranges (private networks, loopback). HTTPS is enforced in production.
  • Timeout protection: A background process monitors for stuck runs and automatically fails them after the configured timeout plus a grace period

​
Callback Endpoint

Your runner reports completion via: 
POST /v1/workflow-runs/{run_id}/callback
The callback token is validated to ensure it matches the specific run.

​
Scheduling Workflows

Knowledge Stack does not include a built-in scheduler. You can trigger workflows on a schedule using any external scheduler (cron, cloud schedulers, CI/CD pipelines): 
# Example: Run every weekday at 9am
0 9 * * MON-FRI curl -X POST \
  https://your-instance.com/v1/workflow-definitions/{id}/invoke \
  -H "Authorization: Bearer sk-user-..."

​
Authorization

OperationWho Can Do It
Create a workflowOwners and admins
View workflowsAnyone with read access to the workflow’s path
Update a workflowAnyone with write access, or owners/admins
Delete a workflowOwners and admins only
Invoke a workflowAnyone with read access to the workflow and appropriate access to all ABCD paths
Delete a runThe user who created it, or owners/admins. Running workflows cannot be deleted.