Proactive Knowledge Graph building: FAQ
Proactive Knowledge Graph building is SIGNLD mapping and proposing structure from a connected system as soon as it connects, instead of learning that structure gradually from chat and queries. Proactive Knowledge Graph building is in beta in SIGNLD as of October 2026, with general availability planned for early 2027.
Related reading in this series: Proactive graph building for companies with no data team and Guardrails: what SIGNLD asks you to confirm before it commits.
For the wider context, see our explainer on what a proactive knowledge graph is.
Key takeaways
- Proactive Knowledge Graph building is a mode in SIGNLD where, once a system connects, SIGNLD inspects its structure and proposes entities and relationships right away, rather than waiting for someone to ask questions that would eventually teach the graph the same thing.
- The main new task is reviewing a queue of proposed entities and relationships, particularly ones with a lower confidence signal, and confirming or correcting them.
- Systems connected before proactive building rolled out continue to have their graph structure refined the way it always was, through usage, chat, and queries.
- Because a proposed match, even a well-supported one, is still a guess about what two records represent, and leadership decisions get made on top of that guess.
- Start with what is a proactive knowledge graph for the category definition, and see the roadmap for what SIGNLD has planned as the feature moves from beta toward general availability.
Frequently asked questions
What is proactive Knowledge Graph building?
Proactive Knowledge Graph building is a mode in SIGNLD where, once a system connects, SIGNLD inspects its structure and proposes entities and relationships right away, rather than waiting for someone to ask questions that would eventually teach the graph the same thing. It produces a first draft earlier, not a finished graph. Every proposal still needs a person to confirm it before SIGNLD treats it as settled.
How is this different from how SIGNLD built the graph before?
Before, the Knowledge Graph grew mainly from usage: chat interactions, data connections, and the questions a team asked over time gradually taught SIGNLD which records related to which. Proactive building adds a second path, where the graph also drafts structure directly from a newly connected system's schema, without needing a question to trigger it. Both paths still route proposals through the same human review step.
Does proactive graph building mean the graph is finished once a system connects?
No. SIGNLD never claims the Knowledge Graph is complete or correct without review, proactively built or not. Connecting a system produces a set of proposed entities and relationships based on its structure. A person still confirms, corrects, or dismisses those proposals before SIGNLD treats them as part of the working graph.
Does proactive building make onboarding faster?
It changes when graph structure appears, not the two approved timing figures: connecting a system still takes about 15 minutes, and a first answer still comes back in minutes. What proactive building adds is a broader draft of the graph earlier in the process, along with a review step for that draft that did not exist as early under the reactive model.
What does a person actually have to do differently with proactive building on?
The main new task is reviewing a queue of proposed entities and relationships, particularly ones with a lower confidence signal, and confirming or correcting them. This is in addition to, not instead of, asking questions and reading Decision Briefs. The honest tradeoff is that proactive building produces a wider draft sooner, and that draft needs a human check before anyone should rely on it.
Does proactive graph building require new permissions from connected systems?
No. It uses the same read-only connection every SIGNLD integration already requires. Proactive building reads the schema and records a connection is already permitted to access; it does not request write access or expand the scope of what a connector can read from a source system.
Where does the AI that proposes graph structure run?
On a single-tenant AWS Bedrock instance dedicated to your tenant, under SOC 2 Type II controls, with tenant isolation enforced between customers. SIGNLD is never trained on your data. This applies to structure proposed proactively from a new connection in the same way it applies to structure learned reactively from usage.
Can SIGNLD commit a proposed relationship to the graph without anyone checking it?
No. Proposed entities and relationships, especially lower-confidence matches, are flagged for review rather than written into the graph as settled fact. SIGNLD proposes; a person decides what the graph treats as true. This applies across the platform, not only to proactive building.
Does SIGNLD claim zero data egress when it builds the graph proactively?
No. SIGNLD does not claim zero egress or that data never leaves your environment, for proactive graph building or any other part of the platform. Read-only source access, tenant isolation, and single-tenant inference are the controls that apply. Full detail is on the security page.
Is proactive graph building available to all SIGNLD customers now?
It is in beta as of October 2026, with general availability planned for early 2027. Customers onboarding or using SIGNLD during the beta period may see proposed graph structure from newly connected systems, alongside the existing reactive learning behavior the platform has always used.
What happens to a system that was connected before proactive building existed?
Systems connected before proactive building rolled out continue to have their graph structure refined the way it always was, through usage, chat, and queries. Proactive building applies its upfront proposal step to systems connected while the feature is active for a tenant; it does not retroactively rescan every historical connection on its own.
Does proactive building change how a Decision Brief is structured?
No. A Decision Brief still has four parts: the finding, the evidence linked back to source records, a confidence score, and a recommended action. Proactive graph building affects how the underlying Knowledge Graph gets its structure, not the format or contents of the brief that gets returned when someone asks a question.
Why does SIGNLD still require review if the system can propose relationships on its own?
Because a proposed match, even a well-supported one, is still a guess about what two records represent, and leadership decisions get made on top of that guess. SIGNLD treats speed and correctness as separate problems: proactive building solves for how fast a draft appears, and human review solves for whether that draft is actually right. Removing the review step would remove the only check on the second problem.
Does proactive graph building work for teams without a dedicated data team?
Yes, and it is often more useful there, because those teams are less likely to know in advance which questions would have taught the graph reactively. Proactive building gives them a draft to react to instead of a blank graph to interrogate. It still requires someone, technical or not, to spend time confirming what gets proposed before relying on it.
Where can I read more about how proactive graph building works?
Start with what is a proactive knowledge graph for the category definition, and see the roadmap for what SIGNLD has planned as the feature moves from beta toward general availability. How it works covers the mechanics of a Decision Brief in full.