Talk to your PhoenixAI data in plain English. Generate, fix, and optimize SQL without leaving the editor. Keep every byte of customer data inside your VPC.
We’re excited to introduce Agent Fawkes, the new AI Assistant built into PhoenixAI Cloud. Agent Fawkes turns the SQl Editor console into a conversational workspace so users can simply ask for SQL and get it written, fix broken queries with a single click, create synthetic tables, schema and data, surface answers from the PhoenixAI docs, and chart your results, all without leaving the page.
PhoenixAI is the fastest SQL engine for real-time and AI-driven analytics on the lakehouse. Trusted by Pinterest, Tencent, Expedia, and more.
Why Agent Fawkes — and why now
Real-time analytics platforms are getting more powerful, but also harder to use. Schemas grow, query syntax gets sharper, and the team writing queries widens beyond data engineers. At the same time, LLMs have reached the point where they can draft good SQL, explain complex query plans, and ground answers in product documentation reliably.
The question for a data platform vendor isn’t whether to ship an AI assistant. It’s how to ship one that’s actually useful, actually safe, and actually fast — especially in a BYOC context, where customer data sits inside the customer’s own VPC.
That’s the brief Agent Fawkes was built to answer.
What Agent Fawkes can do today
- Natural-language to SQL. Ask “how many active users did we have last week?” and get a draft query, ready to run.
- One-click SQL fix. Failed queries surface a Fix button that returns a corrected version with a short reason.
- Optimize, rewrite, explain. Slow query? Get an optimized variant. Gnarly CTE? Get a plain-English walkthrough. Need it in a different style? Get a clean rewrite.
- Auto-suggest charts. Run a query, click Chart, and Fawkes proposes a sensible visualization — without ever seeing the underlying rows.
- Documentation Q&A. Ask configuration, feature, or troubleshooting questions and get answers grounded in the official PhoenixAI docs, with clickable citations.
- Schema browsing in chat. “List my tables.” “Show me the create-table for orders.” “How is my routine load doing?” Answers come from your live cluster, not stale snapshots.
Streaming responses, conversation memory inside a thread, and instant cancel keep the experience fluid even as queries get more complex.
Built for the BYOC trust model
Agent Fawkes was engineered with the same principles that drove BYOC itself: keep customer data inside customer infrastructure, give organizations real control over what’s enabled, never trade safety for convenience.
- Row data never reaches the LLM. Chart generation sees only column names, types, and statistics, never actual cells.
- Credentials never leave the region. The agent holds an opaque driver ID; the real PhoenixAI login lives in a region-local connection pool inside your environment.
- Conversation history is encrypted at rest. Per-cluster KMS keys, rotated every 30 minutes. A raw DB dump is useless without the central key.
- Explicit consent for cluster access. Every chat asks before reading your schema. Allow once or Allow for this session.
- Org-level enable/disable. Account or org admins decide whether Fawkes is available, and can turn it off cluster-wide at any time.
- Tenant tags flow end-to-end. Every LLM call is attributable to an account, cluster, and region. no shared-pool ambiguity.
Who benefits
- SQL writers and analysts spend less time wrestling syntax and more time on the question that actually matters.
- New PhoenixAI users get a guided ramp from “what’s a tablet?” to running their first production query, without ever leaving the console.
- Operators and platform teams can ask Fawkes how a load job is doing, why a query was slow, or which tables haven’t been touched recently — all in chat.
- Security and compliance reviewers get a system that was designed around their concerns, not retrofitted to them.
The Real-Time Analytics Benchmark Report 2026
P99 latency, concurrency, and cost benchmarks — PhoenixAI vs. ClickHouse, Druid, and Trino.
What’s next on the roadmap
Agent Fawkes is the foundation, not the finish line. Coming next:
- Native vector storage and search inside Phoenix — store and retrieve embeddings alongside your operational data.
- MCP server for external agents — let Claude, OpenAI, or your internal tools talk to PhoenixAI directly.
- Inline LLM SQL functions like prompt() and call_llm() — add intelligence right inside a query.
- Semantic-layer auto-generation — let Fawkes propose canonical metric and dimension definitions for downstream agents to reuse.
- Customer-managed LLMs — for organizations that prefer to keep inference inside their own infrastructure.
Quick Demo on Agent Fawkes
Get Started Today
Agent Fawkes is rolling out now to PhoenixAI Cloud customers, with no extra cost during the preview period.
- Visit the PhoenixAI Cloud console at https://cloud.PhoenixAI.com/
- Have your account or org admin enable the AI Assistant from Account → Settings
- Open the chat panel and ask Fawkes anything
We’d love your feedback — every response carries a thumbs-up / thumbs-down button, and your reactions go straight to the team building the next version.