agent.tools v1.0.0 Tool Agent
LLM agent with iterative tool calling.
- In
- Prompt
- Out
- Output · Tool Calls · Usage
Operator catalog
PublicEvery operator declared under /opt/glyph/schemas/operators/ is listed below. The page is auto-generated at build time from the JSON schemas, so the catalog cannot disagree with the runtime registry.
For deeper per-family reference pages — config tables, port lists, source schema paths — see Operators by family. The same JSON schemas drive both views; the family pages are emitted by docs/scripts/gen-operator-catalog.mjs.
For the conceptual story behind operators, see Why operators (not custom code).
Agent 1
Cache 1
cache.entry v1.0.0 Cache
Caches step results by key with configurable TTL.
- In
- Input
- Out
- Output · Cache Hit
Code 1
code.sandbox v1.0.0 Code Sandbox
Evaluates CEL expressions or JavaScript in a sandboxed environment.
- In
- Input
- Out
- Output
Connector 4
connector.action v1.0.0 Connector Action
Executes an action on a connected external service.
- In
- Input
- Out
- Output
connector.invoke v1.0.0 Connector Invoke
Calls a third-party API through a tenant-scoped connector credential (OAuth2, API key, basic, bearer).
- In
- Input
- Out
- Output
connector.list_resources v1.0.0 Connector List Resources
Lists resources of a given kind from a connector (e.g. customers, charges). Dispatched by connector_kind to per-kind handlers.
- In
- Input
- Out
- Output
connector.search v1.0.0 Connector Search
Searches a connected external service.
- In
- Query
- Out
- Results
Data 3
data.list v1.0.0 List Operations
Operation-rich list manipulation: map, filter, sort, chunk, flatten, dedupe, aggregate, split.
- In
- Input
- Out
- Output
data.object v1.0.0 Object Operations
Operation-rich object manipulation: assign, select, rename, merge.
- In
- Input
- Out
- Output
data.parse v1.0.0 Parse / Stringify
Converts between formats: JSON, CSV, markdown table, URL-encoded, plain text.
- In
- Input
- Out
- Output
Database 2
db.mutate v1.0.0 Database Mutate
Executes one or more INSERT/UPDATE/DELETE statements as a single transaction under tenant RLS. Each statement is allowlisted (no DDL) and parameterised. Writes an audit row to glyph.operator_audit on commit.
- In
- Input
- Out
- Output
db.query v1.0.0 Database Query
Executes a parameterized SQL query.
- In
- Input
- Out
- Rows · Row Count
Document 4
document.clone v1.0.0 Document Clone
Duplicates a source document into a new document row, copying all four subdoc snapshots (graph, draw, meta, comments). Defaults to the current run's document. Collaborators are not copied.
- In
- none
- Out
- Cloned Document
document.extract v1.0.0 Document Extract
Extracts text, tables, or metadata from documents (PDF, DOCX, etc.).
- In
- File
- Out
- Text · Metadata
document.list_collaborators v1.0.0 Document Collaborators
Returns the collaborator roster for a document with their permission bits. Defaults to the current run's document. The email and name fields are surfaced as null — the collaborator row carries only the opaque user id.
- In
- none
- Out
- Collaborators
document.publish v1.0.0 Document Publish
Captures all four subdocs of the target document (graph, draw, meta, comments) into a new revision row, marks the revision as published, and sets it as the document's published_revision_id.
- In
- none
- Out
- Published Revision
Eval 4
eval.assertion v1.0.0 Assert (CEL)
Evaluates a CEL boolean expression against merged inputs and fails the step if it returns false.
- In
- Input
- Out
- Output
eval.grade v1.0.0 LLM Grader
Scores a response against a rubric using an LLM judge. Returns a score, justification, and per-criterion results.
- In
- Input
- Out
- Output
eval.regex v1.0.0 Regex
Native regex match/find_all/replace against a target string.
- In
- Input
- Out
- Output
eval.score v1.0.0 Evaluate & Score
Scores output quality against a reference or rubric.
- In
- Input
- Out
- Score · Explanation · Usage
Guardrails 4
guardrails.check v1.0.0 Guardrails Check
Validates text against safety/compliance rules.
- In
- Input
- Out
- Pass · Violations · Usage
guardrails.cost_cap v1.0.0 Cost Cap
Compares the run's accumulated USD spend against a hard cap. Fails or throttles when exceeded.
- In
- Input
- Out
- Output
guardrails.input_filter v1.0.0 Input Filter
Scans an input string for PII / prompt injection / jailbreak patterns and can fail, redact, or flag.
- In
- Input
- Out
- Output
guardrails.output_filter v1.0.0 Output Filter
Validates an output against a JSON schema, truncates by max_tokens, and (optionally) runs an external toxicity check.
- In
- Input
- Out
- Output
HTTP 1
http.request v1.0.0 HTTP Request
Makes an HTTP request to an external URL.
- In
- Input
- Out
- Response
Human 2
human.approval v1.0.0 Human Approval
Pauses the run and waits for human approval before continuing.
- In
- Input
- Out
- Approved · Rejected
human.input v1.0.0 Human Input
Pauses the run and requests manual data entry.
- In
- Input
- Out
- Output · Timed Out
Input 5
input.file v1.0.0 File Input
References a file from storage.
- In
- none
- Out
- Output
input.json v1.0.0 JSON Input
Provides a static JSON value.
- In
- none
- Out
- Output
input.sample_data v1.0.0 Sample Data
Provides sample input data for testing workflows.
- In
- none
- Out
- Output
input.secret v1.0.0 Secret Reference
Binds a named secret for use downstream.
- In
- none
- Out
- Output
input.text v1.0.0 Text Input
Provides a static or template text value.
- In
- none
- Out
- Output
Knowledge 4
knowledge.delete v1.0.0 Knowledge Delete
Removes chunks from a knowledge namespace, optionally scoped to a single doc_key.
- In
- Input
- Out
- Output
knowledge.list v1.0.0 Knowledge List
Paginates chunk metadata for a knowledge namespace.
- In
- Input
- Out
- Output
knowledge.search v1.0.0 Knowledge Search
Semantic search over a pgvector knowledge namespace via cosine similarity.
- In
- Query
- Out
- Output · Results · Scores
knowledge.upsert v1.0.0 Knowledge Upsert
Chunks, embeds, and persists text into a pgvector-backed knowledge base.
- In
- Content
- Out
- Output
Logic 7
logic.if v1.0.0 If / Else
Routes data based on a CEL condition.
- In
- Input
- Out
- True · False
logic.iterate v1.0.0 Iterate / For Each
Fans out: executes downstream once per item in input array.
- In
- Input
- Out
- Item · Results
logic.join v1.0.0 Join / Parallel
Waits for all upstream branches before continuing.
- In
- Branch A · Branch B
- Out
- Output
logic.loop v1.0.0 Loop
Repeats a subgraph until a condition is met or max iterations reached.
- In
- Input
- Out
- Output · Iteration
logic.switch v1.0.0 Switch
Routes data to one of N branches based on a CEL expression.
- In
- Input
- Out
- Default
logic.wait v1.0.0 Wait / Delay
Pauses execution for a specified duration.
- In
- Input
- Out
- Output
logic.wait_for_event v1.0.0 Wait for Event
Suspends until a matching external event or timeout.
- In
- Input
- Out
- Event Payload · Timed Out
Media 5
media.generate v1.0.0 Generate Image
Generates an image from a text prompt via OpenRouter (or any OpenAI-compatible image endpoint). Persists the image to tenant storage.
- In
- Input
- Out
- Image
media.ocr v1.0.0 OCR
Extracts text from an image using a vision-capable LLM. Counts against the tenant AI budget.
- In
- Input
- Out
- Text
media.process v1.0.0 Media Process
Processes images, audio, or video (resize, transcribe, extract frames).
- In
- Media
- Out
- Output
media.transcribe v1.0.0 Transcribe Audio
Transcribes audio to text via a Whisper-compatible endpoint. Counts against the tenant AI budget.
- In
- Input
- Out
- Transcript
media.transform v1.0.0 Transform Image
Applies local image transformations (resize, re-format, strip metadata). No external API call.
- In
- Input
- Out
- Image
Model 5
model.classify v1.0.0 Classify
Classifies text into predefined categories using an LLM.
- In
- Input
- Out
- Classification · Usage
model.embed v1.0.0 Embed
Generates vector embeddings from text.
- In
- Input
- Out
- Embedding · Usage
model.extract v1.0.0 Extract Structured
Extracts structured data from text using an LLM with a JSON schema.
- In
- Input
- Out
- Extracted · Usage
model.rerank v1.0.0 Rerank
Reranks documents by relevance to a query.
- In
- Query · Documents
- Out
- Ranked
model.respond v1.0.0 Respond
Sends a prompt to an LLM and returns generated text.
- In
- Prompt
- Out
- Text · Usage
Output 5
output.export v1.0.0 File Export
Exports output as a downloadable file.
- In
- Input
- Out
- Artifact
output.notify v1.0.0 Notification
Sends a notification via email, Slack, webhook, etc.
- In
- Input
- Out
- none
output.persist v1.0.0 Persist
Writes output data to a persistent store.
- In
- Input
- Out
- Output
output.preview v1.0.0 Preview Output
Displays output in the run console for debugging.
- In
- Input
- Out
- none
output.return v1.0.0 Return Output
Returns the final workflow output.
- In
- Input
- Out
- none
Prompt 2
prompt.messages v1.0.0 Chat Messages
Constructs a multi-turn chat message array for LLM consumption.
- In
- Dynamic Content
- Out
- Messages
prompt.template v1.0.0 Prompt Template
Renders a prompt string from a Handlebars/Jinja-style template.
- In
- Variables
- Out
- Output
Storage 1
storage.object v1.0.0 Storage Object
Reads or writes objects in S3-compatible storage.
- In
- Input
- Out
- Output
Tools 3
tools.call v1.0.0 Call Tool
Invokes one of the built-in tools (calculator, datetime, regex_match, json_path, http_fetch).
- In
- Input
- Out
- Output
tools.list v1.0.0 List Tools
Returns the static catalog of built-in tools and their argument schemas.
- In
- none
- Out
- Output
tools.mcp v1.0.0 MCP Tools
Connects to an MCP server and exposes its tools.
- In
- Input
- Out
- Output · Tool Results
Trigger 7
trigger.chat v1.0.0 Chat Trigger
Starts a run from a chat message.
- In
- none
- Out
- Message · History
trigger.connector v1.0.0 Connector Trigger
Starts a run on an external connector event.
- In
- none
- Out
- Event
trigger.form v1.0.0 Form Trigger
Starts a run from a submitted form.
- In
- none
- Out
- Form Data
trigger.manual v1.0.0 Manual Trigger
Starts a run when a user clicks Run.
- In
- none
- Out
- Trigger
trigger.schedule v1.0.0 Schedule Trigger
Starts a run on a cron or interval schedule.
- In
- none
- Out
- Trigger
trigger.webhook v1.0.0 Webhook Trigger
Starts a run when an HTTP webhook is received.
- In
- none
- Out
- Payload · Headers
trigger.workflow v1.0.0 Workflow Trigger
Starts a run when called by another workflow.
- In
- none
- Out
- Input
Vars 1
vars.state v1.0.0 Workflow State
Reads or writes persistent workflow-scoped state variables.
- In
- Input
- Out
- Output
Web 3
web.fetch v1.0.0 Web Fetch
Fetches content from a URL.
- In
- URL
- Out
- Content · Metadata
web.scrape v1.0.0 Web Scrape
Fetches a URL through the SSRF-guarded HTTP client and extracts the main article body in markdown / text / html, plus lightweight metadata (author, description, published, site).
- In
- URL
- Out
- Article
web.search v1.0.0 Web Search
Searches the web using a search API.
- In
- Query
- Out
- Results
Workflow 3
workflow.call v1.0.0 Call Workflow
Calls another published workflow as a child run.
- In
- Input
- Out
- Output
workflow.invoke v1.0.0 Invoke Workflow
Invokes another deployment as a child run with depth-limited recursion. Calls the engine directly (no HTTP self-loopback) and enforces customer scoping at both the SQL and handler layer.
- In
- Input
- Out
- Output
workflow.return v1.0.0 Return From Workflow
Terminates the current run early with an explicit value. The DAG scheduler detects the terminal marker on this step's output and finalises the run as completed, skipping any remaining downstream steps.
- In
- Input
- Out
- Output