Kylon vs. PromptQL: Query layer vs. work layer

Hasura's PromptQL is one of the most technically rigorous AI agent platforms on the market. While most AI tools generate answers and hope they're right, PromptQL takes a radically different approach: the AI writes a step-by-step query plan, then executes it deterministically in Python and SQL — outside the model. No hallucinations. Every step inspectable. Every result verifiable.

For data teams, that's exactly what they've been asking for. But for the rest of the organization? That's where the story gets more nuanced.

What PromptQL does well

Let's start with what makes PromptQL genuinely excellent.

The deterministic execution model is a breakthrough for enterprise data work. Traditional AI agents generate answers from a language model — and sometimes those answers are confidently wrong. PromptQL's architecture separates thinking from executing: the AI reasons about how to get the answer, then executes the plan in real code. You can inspect every step, every query, every intermediate result. For financial services, healthcare, and any domain where accuracy isn't optional, this matters enormously.

Cross-system data joins are a killer feature. Built on Hasura's DDN (Data Delivery Network), PromptQL can stitch together data from Snowflake, BigQuery, PostgreSQL, MongoDB, Salesforce, and SaaS APIs — all through one natural-language interface. "Show me customers who opened support tickets last month but whose NPS score increased" isn't a two-hour spreadsheet exercise anymore. It's a single prompt.

The self-building wiki is smart: it seeds itself from Slack, Google Docs, Snowflake, and other sources, maintains revision history, and enforces scoped access. Shared threads let team members collaborate on analyses. Automations can be deployed as HTTP endpoints, turning one-off queries into repeatable workflows.

And the security model is enterprise-grade: row-level and column-level permissions, SOC 2/HIPAA/GDPR/ISO 27001 compliance, role-based access control, and audit logging. For regulated industries, PromptQL checks every box.

Where the query model hits its limits

PromptQL solves data querying better than anything else on the market. But querying data is one part of how teams work.

It's built for data teams. The supergraph concept, the query plan model, the Python/SQL execution — all of it assumes the user thinks in terms of data sources and queries. Your VP of Sales doesn't think in supergraphs. Your marketing lead doesn't want to inspect Python execution plans. PromptQL is a power tool for analysts and data engineers, and it's honest about that.

Agents analyze — they don't act. PromptQL agents can find the answer to any question across your data stack. But they can't send the follow-up email. They can't update the CRM record. They can't deploy a landing page, schedule a social post, or build an internal dashboard. The agent's job ends at the query result.

Collaboration is thread-based. PromptQL's "multiplayer" means shared threads and a shared wiki. That's useful for data team collaboration. But it's not a workspace where marketing, sales, ops, and engineering all work together with AI agents across their daily workflows.

How Kylon approaches this differently

A work platform, not a query platform

PromptQL answers questions. Kylon does work.

When a Kylon agent receives a question like "show me underperforming campaigns," it doesn't just pull the data — it can also draft a remediation plan, update the campaign database, notify the team lead, and schedule a check-in for next week. The query is the beginning of the workflow, not the end.

Kylon agents send emails, update databases, deploy pages, manage project pipelines, process documents, monitor KPIs, and coordinate with other agents — all within the same workspace. The data layer is part of it, but execution is the point.

For the whole team, not just analysts

PromptQL assumes its users are comfortable with data concepts. Kylon assumes its users are everyone on the team.

A sales rep asks an agent to update the pipeline. A marketing lead asks an agent to draft a blog post and deploy it. An ops manager asks an agent to build a project tracker. A CEO asks an agent for a weekly summary. None of these people need to think about query plans or supergraphs. They describe what they need in plain language, and agents handle the execution.

This isn't about dumbing things down. It's about matching the tool to the full range of work a team does — not just the data analysis slice.

Agents that take action

The deepest difference is in what happens after the analysis.

In PromptQL, the workflow ends with a result: a chart, a CSV, a summary. What you do with that result is up to you. Copy it into a Slack message. Paste it into a presentation. Manually update the relevant system.

In Kylon, agents close the loop. The agent finds underperforming campaigns, updates their status in the built-in database, notifies the relevant team member with a structured card, and schedules a follow-up workflow to check results in a week. The human reviews and approves. The agent executes. No copy-paste. No manual handoff.

Multi-agent coordination

PromptQL is one AI layer focused on data. Kylon runs multiple specialized agents — a data agent, a content agent, a project agent, an ops agent — all working in the same workspace, sharing context, handing off tasks, and collaborating with each other and with humans.

When the data agent finds an interesting trend, the content agent can draft a report about it. When the ops agent detects an anomaly, the project agent creates a task. The coordination happens naturally because all agents share the same workspace context.

When each tool makes sense

Choose PromptQL when your primary challenge is querying data across multiple systems with guaranteed accuracy. If you're a data team, analytics organization, or any team in a regulated industry where "close enough" isn't good enough — PromptQL's deterministic execution model is best-in-class. It's the most reliable way to get truthful answers from complex data.

Choose Kylon when your team needs AI agents that don't just answer questions but do the work. If your challenge spans data analysis, content creation, project management, customer operations, and team coordination — and you want agents that act, not just report — Kylon is the platform built for that.

Query vs. work

The data query problem and the team work problem are different problems. PromptQL solves the first one brilliantly. Kylon solves the second one. The question for your team is: which bottleneck are you hitting?

If you're drowning in data but the rest of your workflow is fine, PromptQL is the answer. If you need AI that works across your entire team — not just the data slice — the workspace approach goes further.