Expression Reference

Complete reference for all expression variables available in TARX workflows. Expressions use the {{variable.field}} syntax and are evaluated at runtime when a node executes.

---

Trigger Node

Node type: triggerNode

Expression	Type	Description	Example Value
{{trigger.output}}	string	The trigger's output — main entry point for all workflows	"artificial intelligence"

By Trigger Type

Trigger Type	{{trigger.output}} Value
Manual	The string entered in the run modal
Webhook	Full JSON payload, or extracted field if payload_path is set
Schedule	The default_input configured on the trigger
HTTP	Full request context JSON: {"method": "POST", "headers": {...}, "body": {...}, "query_params": {...}}

---

Agent Node

Node type: agentNode

The expression variable uses the agent's name (from Agent Editor), not the canvas node label.

Expression	Type	Description
{{agent_name.output}}	string	The full text output from the LLM call

Multiple Agents in One Workflow

All previous agents' outputs are available to any downstream node:

# In a third agent node's input:
Research: {{researcher.output}}
Draft: {{writer.output}}
Please review and combine these.

Caveat: Can only reference agents that have already completed (upstream agents). Referencing a downstream or non-existent agent returns empty string.

---

Loop / Iterator Node

Node type: loopNode

Expression	Type	Available In	Description
{{loop_name.current_item}}	string	Inside loop body	Current item being processed
{{loop_name.index}}	int	Inside loop body	0-based index of current iteration
{{loop_name.output}}	string (JSON array)	After loop completes	JSON array of all iteration results

Example Values

For a loop named content_loop iterating over ["topic A", "topic B", "topic C"]:

During iteration 1:

• {{content_loop.current_item}} → "topic A"
• {{content_loop.index}} → 0

During iteration 2:

• {{content_loop.current_item}} → "topic B"
• {{content_loop.index}} → 1

After all iterations:

• {{content_loop.output}} → ["result for A", "result for B", "result for C"]

---

Branch Node

Node type: branchNode

Expression	Type	Description
{{branch_name.output}}	string	"True" or "False" based on condition evaluation

The branch node also exposes output as a plain variable in the condition expression itself (not via {{...}} syntax — it's a Python variable in the condition field).

---

Merge Node

Node type: mergeNode

Expression	Type	Description
{{merge_name.output}}	string	Output from the completed branch (first mode) or combined outputs (all mode)

---

Human-in-Loop Node

Node type: humanLoopNode

Expression	Type	Description
{{hil_name.output}}	string	Decision result including any reviewer comment

HiL Output Format

Decision	Output Value
Approved (no comment)	"approved"
Approved with comment	"approved: [reviewer's comment]"
Rejected with comment	"rejected: [reviewer's comment]"
Timeout auto-reject	"rejected: timeout — approval expired after N hours"

---

Output Node

Node type: outputNode

Output nodes are leaf nodes — they don't expose expressions to other nodes (they're the final collection point). Their value field uses expressions from upstream nodes.

---

Expression Resolution Rules

1. Execution Order

Expressions resolve to empty string "" if the referenced node hasn't completed yet. Nodes can only reference upstream (already-completed) nodes.

# Valid: writer comes after researcher
researcher → writer
writer's input: "{{researcher.output}}"  ✅

# Invalid: researcher references writer which hasn't run yet
writer → researcher
researcher's input: "{{writer.output}}"  → ""  ❌

2. Case Sensitivity

Agent names in expressions are case-sensitive and must exactly match the agent's name as configured in the Agent Editor.

# Agent named: researcher (lowercase)
{{researcher.output}}  ✅ Matches
{{Researcher.output}}  ❌ Does not match (returns "")
{{RESEARCHER.output}}  ❌ Does not match (returns "")

3. Whitespace in Names

Agent names cannot contain spaces (the Agent Editor enforces this). Use hyphens: seo-writer, not seo writer.

{{seo-writer.output}}  ✅ Valid
{{seo writer.output}}  ❌ Invalid syntax

4. Nested Expressions

Expressions cannot be nested:

{{{{agent1.output}}.something}}  ❌ Not supported

Use a Transform node (JSONPath) to extract a field, or have an agent produce the simple string you need for downstream use.

5. Expression in Different Field Types

Field Type	Expression Support
Text input fields	✅ Full expression support
Dropdown selectors	❌ No expression support
Boolean toggles	❌ No expression support
Numeric fields	⚠️ Limited — {{loop.index}} works if the field accepts the resulting numeric string

---

Condition Expression Variables

In Branch node condition fields, these Python variables are available (not {{...}} syntax — these are direct Python variables):

Variable	Type	Value
output	str	Output of the immediately upstream node
trigger_output	str	The original trigger output
node_outputs	dict	All completed node outputs: {"agent_name": "output_text"}

Condition Examples Using Variables

# Basic string check using output
"error" in output.lower()

# Length check
len(output) > 500

# Access specific agent output
node_outputs.get("researcher", "") != ""

# Parse JSON from agent output
__import__('json').loads(output).get("severity") == "critical"

# Cross-reference two agents
node_outputs["reviewer_1"].strip() == node_outputs["reviewer_2"].strip()

# Numeric comparison
float(output.strip()) > 0.8

---

Complete Example

A workflow with all node types, showing which expressions are available at each step:

1. Trigger (Manual): "research quantum computing"
   → {{trigger.output}} = "research quantum computing"

2. Loop: item_loop, items = ["quantum gates", "quantum error correction", "quantum supremacy"]
   During iteration 1:
   → {{item_loop.current_item}} = "quantum gates"
   → {{item_loop.index}} = 0
   
3. Agent: researcher, input = "Research: {{item_loop.current_item}} in context of {{trigger.output}}"
   → {{researcher.output}} = "Quantum gates are the building blocks..."

4. Agent: writer, input = "Write intro based on:\n{{researcher.output}}"
   → {{writer.output}} = "In the world of quantum computing..."

5. Branch: is_long, condition = "len(output) > 500"
   → {{is_long.output}} = "True" or "False"

6. [True path] Agent: summarizer, input = "Summarize:\n{{writer.output}}"
   → {{summarizer.output}} = "A brief summary..."

7. [False path] Agent: expander, input = "Expand to 300 words:\n{{writer.output}}"
   → {{expander.output}} = "Expanded version..."

8. Merge: content_merge
   → {{content_merge.output}} = whichever path ran

9. After loop completes:
   → {{item_loop.output}} = ["result 1", "result 2", "result 3"]

10. Output: {{item_loop.output}}