Canonical: https://molo17.com/spark/ Markdown mirror: https://molo17.com/spark/index.md Title: Spark · The Gluesync AI Agent Layer · MOLO17 Description: Spark is the Gluesync AI agent layer. It works through Core Hub MCP tools, MOLO17 online documentation and knowledge tools, and Gluesync Connect capabilities. Meet Spark · Gluesync AI Hub # The agent layer that knows your estate. Spark is the AI agent layer inside Gluesync. It reasons with your model, acts through governed Core Hub tools, reads MOLO17's own documentation and knowledge tools, and extends across sites with Gluesync Connect. [Book a demo](/book-a-demo/) [Build an agent in AI Studio](/capabilities/ai-studio/) Spark · Agent layer Governed session MCP **Core Hub MCP** _107 tools_ DOC **MOLO17 docs MCP** _Published knowledge_ SITE **Gluesync Connect** _Coming soon_ `spark.listen()` Spark waits with your model, your keys, and one allow-list of tools. `get_pipeline_status("orders-to-analytics")` It inspects pipelines, agents, and metrics with the signed-in user's permissions. `search_documentation("cdc lag tuning")` It answers from MOLO17's published documentation instead of model memory. `summarize_sites(scope: "fleet")` The same agent is designed to reach every site Connect registers. One agent layer · Three tool sources ## Spark is only as good as what it can safely reach. Instead of a model guessing about your environment, Spark calls tools that already exist — in your Core Hub, in MOLO17's published knowledge, and across the sites Connect will manage. 01 ### Core Hub MCP tools Spark reaches pipelines, agents, entities, connections, metrics, notifications, Chronos, and Query Studio SQL through the 107-tool embedded catalog. Each agent carries an explicit allow-list, and every call runs with the signed-in user's Core Hub permissions. [Explore Core Hub MCP →](/corehub-mcp/) 02 ### MOLO17 online MCP tools Spark can also consult MOLO17's own hosted MCP tools for product documentation and knowledge, so an answer about Gluesync behavior cites published material instead of improvising from model memory. [Browse the documentation →](https://docs.molo17.com) 03 ### Gluesync Connect capabilities Connect's hosted assistant is designed to run Spark across many registered sites for fleet-wide questions and incident summaries. Connect is coming soon, planned for Q4 2026, and available today through the waitlist. [See Gluesync Connect →](/connect/) Where Spark shows up ## The same agent, reached four ways. [AI Studio chat ### Ask, investigate, and report Spark agents answer questions about the estate, explain notifications in plain language, and send severity-aware reports through configured webhooks and email. _Explore →_](/capabilities/ai-studio/) [Published runs ### Repeatable, durable work Freeze an agent into an immutable version, then start asynchronous runs with idempotency, cancellation, and an ordered event timeline. _Explore →_](/capabilities/ai-studio/#section-developer-api) [Query Forge SQL ### A Spark reply in a result row Name an agent with spark\_agent in a gluesync.ai query and read its answer, optional SQL draft, citations, and conversation ID as columns. _Explore →_](/query-forge/#sql-callable-ai) [Chronos ### Agents on a schedule or a signal Chronos invokes a published agent through ai\_agent\_run from a cron schedule, a chained pipeline event, or an inbound webhook — with prompt templates, a payload allow-list, and loop guards. _Explore →_](/schedules-and-events/) Why operators allow it ## Useful in production because it stays bounded. Spark inherits the controls your team already trusts, so a helpful agent never becomes an unreviewable one. [Read the AI Studio documentation →](https://docs.molo17.com/gluesync/latest/gs-modules/ai-studio.html) ### The caller's permissions, not the agent's Spark forwards the signed-in identity through existing Core Hub authorization. There is no privileged AI service account behind the conversation. ### Read-only unless you say otherwise SQL through Spark follows Query Studio controls and stays read-only by default. The gluesync.ai path only ever offers read-only tools. ### Writes wait for a human A pipeline mutation never runs straight from the model. The first call is held as a pending action bound to owner, conversation, and pipeline until someone confirms it. ### Traces without secrets Every turn shows which tools ran and how they finished. Provider keys, connection credentials, and raw result rows stay out of the trace. Your infrastructure · your models · your permissions ## Put Spark to work on the estate you already run. [Book a demo](/book-a-demo/) [Explore AI Hub](/ai-hub/)