Developers · Templates

Template Authoring

Templates are pre-built pipelines packaged as JSON. When deployed, Pulse checks prerequisites, runs a wizard to collect configuration, substitutes placeholders, and wires every node into a running agent graph.

Back to Developer Hub

Where templates live

Built-in

Shipped with the release. Curated and reviewed before merge. Available to every install.

User templates

Per-install, hot-reloaded on demand. Drop a JSON file into the user-templates folder of your data directory; it never propagates to other installs.

Template anatomy

A template is a single JSON document. Top-level metadata, a prerequisites list, the wizard's config fields, the node graph, and an optional topics manifest.

json

1{
2  "id": "job-matcher-auto-apply",
3  "name": "Job Matcher, Auto Apply",
4  "description": "Polls job sources, scores against the candidate CV, drafts tailored applications, routes through optional human review, and dispatches.",
5  "category": "Human Resources",
6  "industry": "HR",
7  "workflow": "routing",
8  "icon": "Briefcase",
9  "tags": ["job", "candidate", "auto-apply"],
10
11  "prerequisites": [
12    { "title": "LLM Provider", "description": "A configured language-model provider." },
13    { "title": "Job Sources", "description": "At least one job-board connector installed." }
14  ],
15
16  "configFields": [
17    {
18      "key": "candidateName",
19      "label": "Your full name",
20      "type": "TEXT",
21      "required": true,
22      "group": "Identity"
23    },
24    {
25      "key": "minScore",
26      "label": "Minimum match score (0-100)",
27      "type": "NUMBER",
28      "required": false,
29      "defaultValue": 70,
30      "group": "Matching"
31    }
32  ],
33
34  "nodes": [
35    {
36      "id": "fetcher",
37      "type": "agent",
38      "label": "Job Fetcher",
39      "engine": "external-tools",
40      "inputTopic": "schedule.jobs-poll",
41      "outputTopic": "raw-jobs",
42      "externalTools": ["jobs.list_new"]
43    },
44    {
45      "id": "scorer",
46      "type": "agent",
47      "label": "Match Scorer",
48      "engine": "llm",
49      "inputTopic": "raw-jobs",
50      "outputTopic": "scored-jobs",
51      "systemPrompt": "Evaluate jobs for {{candidateName}}. Score 0-100 based on skills match."
52    }
53  ],
54
55  "topics": [
56    { "name": "raw-jobs", "description": "Jobs as received from sources, pre-scoring." }
57  ]
58}

Config fields, 8 types

configFields drives the deploy wizard. Each entry sets the input control and how the value is substituted into nodes.

Type	Renders as	Substitution
TEXT	Single-line input	{{key}} → verbatim
EMAIL	Email input	{{key}} → verbatim
PASSWORD	Masked input	{{key}} → never logged
NUMBER	Numeric input	{{key}} → as number
TOGGLE	Switch	{{key}} → "true" / "false"
SELECT	Dropdown	{{key}} → option value
TAGS	Chip input	{{key}} → CSV string
UPLOAD	File dropzone	{{key}} → file path

SELECT example

json

1{
2  "key": "llmProvider",
3  "label": "LLM provider",
4  "type": "SELECT",
5  "options": [
6    { "value": "local",     "label": "Local model" },
7    { "value": "managed-a", "label": "Managed provider A" },
8    { "value": "managed-b", "label": "Managed provider B" }
9  ],
10  "defaultValue": "local"
11}

Conditional visibility

json

1{
2  "key": "webhookUrl",
3  "label": "Webhook URL",
4  "type": "TEXT",
5  "visibleWhen": { "field": "dispatcher", "equals": "webhook" }
6}

Placeholder substitution

Dotted paths work: {{candidate.name}}.
List values become CSV.
Booleans become "true" / "false" strings.
Missing keys fail-fast in the wizard.
Substitution is recursive, resolved left-to-right.

Before

json

"systemPrompt": "Hi {{candidateName}}. Match jobs scoring at least {{minScore}}. Required skills: {{keywords}}."

After wizard fill

json

"systemPrompt": "Hi Alice. Match jobs scoring at least 75. Required skills: kubernetes,terraform."

Prerequisites

Two flavours: declarative checks Pulse evaluates automatically (plugin installed, setting non-blank), and human-readable prose for ambiguous setups (pick-one-of-many, version-sensitive).

Most templates mix both, declarative for the basics, prose for the nuances.

Node engines

rule-based

Deterministic routing on field comparisons. No model calls.

json

1{
2  "engine": "rule-based",
3  "rules": [
4    { "condition": "score >= {{minScore}}", "action": "emit", "target": "qualified" },
5    { "condition": "always",                 "action": "emit", "target": "rejected" }
6  ]
7}

llm

Reasoning agent backed by a language model. Optional consensus mode adds redundancy on critical steps.

json

1{
2  "engine": "llm",
3  "systemPrompt": "...",
4  "temperature": 0.3,
5  "maxTokens": 1000,
6  "consensus": { "enabled": true, "quorumSize": 3 }
7}

external-tools

Scripted tool-calling. Executes tools in prompt order without inter-step reasoning. For branching, use llm.

json

1{
2  "engine": "external-tools",
3  "externalTools": ["jobs.list_new", "jobs.list_recent"],
4  "systemPrompt": "Use the tools to fetch new postings every cycle.",
5  "maxToolCalls": 5
6}

Best practices

IDs are permanent

Changing an id breaks every deployed instance. Use lowercase-kebab-case and pick carefully.

Strong descriptions

The root description drives adoption. Name workflows explicitly; describe the output.

Group fields by narrative

Identity → Preferences → Credentials → Advanced. Narrative beats alphabetical.

Defensive defaults

Every optional field needs a sane default. Deploy-without-fill must yield a working pipeline.

Validate in the wizard

Fail-fast in the wizard on missing credentials or invalid ranges, not at the third event in production.

Name topics semantically

raw-jobs beats topic-1. Operators read these names in the Flow page.

Opt into consensus

Don't default every llm node to consensus mode, only where a single-model mistake is expensive.

Pass-through fields

Use passThroughFields to preserve event state through agents that only modify a subset.

Test cases

Embed test cases inside the template; the dry-run endpoint replays them against the deployed graph during development.

json

1"testCases": [
2  {
3    "name": "matches well-fitting job",
4    "input":  { "title": "Senior Rust Developer", "salary": 180000 },
5    "expectedOutput": { "score": { "min": 80 } }
6  }
7]

Shadow templates

Mark a template as shadow-only. It only appears under Pipelines → Shadows. Operators run it against real traffic, compare to production, then promote when confident.

json

1{
2  "id": "job-matcher-v2-experimental",
3  "shadow": true,
4  "nodes": [...]
5}

Publishing

Built-in (PR)

Add the JSON to the pipeline-templates resources, run the lint-audit test suite, submit a PR with a short justification.

User template

Drop the JSON into the user-templates folder of your data directory and reload from Templates → Refresh. Appears with a Custom badge; never shared automatically.

Troubleshooting

Symptom	Fix
Template not found	Verify the file is under the right folder with a .json extension.
Unknown placeholder {{foo}}	Add foo to configFields or check the spelling.
Missing required prerequisite	The referenced plugin isn't installed, fix in Settings.
Agent never receives events	Confirm inputTopic matches the upstream agent's outputTopic.
Validation failed: schema mismatch	Events don't match the topic schema, adjust schema or producer.

Browse the template gallery

See the published catalogue of pipelines you can deploy in one click.

Open templates