Databend @DatabendLabs

Agent performance is not just model performance. Same task: Claude Code + Opus 4.6: ~3 min Claude Code + DeepSeek V4 Pro: ~15 min The difference is not only the model. It is model + harness fit. Tool names, prompts, planning behavior, and runtime conventions shape the path. Final pass/fail won’t show this. The trace will. That’s why Databend is built for agent trace storage, search, and incremental aggregation. Get free credits and try Databend Cloud: app.databend.com/login #AIEngineering #DataArchitecture

Agent traces are not traditional observability traces. They are longer, messier, nested, schema-shifting, and full of tool calls, model outputs, token cost, and branch points. That changes the data layer: object storage, JSON, search, governance, incremental aggregation. That’s what Databend is built for. bit.ly/4vsT0ye #AgenticAI #DataEngineering

Most warehouses let you mask a column. But what about a single key inside a JSON column? Databend's VARIANT sub-field masking lets you selectively hide JSON keys (like secret_key, token) based on role — while the rest of the object stays queryable. Why this matters for AI observability & eval systems (like Langfuse, LangSmith) → One JSON column, multiple sensitivity levels → Role-based visibility at the key level, not just column level → All access paths respect the mask: subscript, json_path_query, cast, json_object_keys No data duplication. No schema redesign. Just policy. docs.databend.com/guides/securit…

How does Databend Cloud scale on 1TB of TPC-H data? • Small: 1173s • Medium: 538s (2.18x faster) • Large: 286s (4.10x faster) 6 billion rows, 22 analytical queries, predictable performance scaling docs.databend.com/guides/benchma…

We just shipped llms.txt for Databend — a progressive skill system for your AI agent. Fetch docs.databend.com/llms.txt for Databend context One line. Your agent gets a structured index of everything Databend — SQL, data loading, tuning, troubleshooting. From there, it progressively fetches deeper docs (sql.txt, guides.txt, tutorials.txt) only when needed. No token waste. Pair it with our MCP server and your agent goes from knowing Databend to acting on your data — query, explore, build — all safely sandboxed. → docs.databend.com/llms.txt → databend.com/mcp/

New in #Databend: Geospatial goes deeper 🌍 - Geometry aggregate funcs in SQL - Refreshable Spatial Indexes - ST_DWITHIN-powered index pruning - Geo values encoded properly in Arrow results Less full-scan geometry. More pruning. Faster spatial queries. Built in Rust, running on your object storage. github.com/databendlabs/d…

New in #Databend: READ_FILE Load any file from a stage directly in SQL. ```sql SELECT READ_FILE('@stage', file_path) FROM documents; ``` Images. PDFs. Binaries. No pipeline. No preprocessing. Just SQL. Docs: docs.databend.com/sql/sql-functi…

We just shipped .md support across all Databend Docs ⚡️ Every guide is now available as clean, raw Markdown — perfect for AI agents, RAG pipelines, and LLMs. No more HTML parsing. Just add .md and go. Try it: → docs.databend.com/guides/ai-func… → docs.databend.com/guides/ai-func… AI devs building on Databend, this one’s for you.

Unlock the power of Databend's new async read_file function! 📂 Read any file from object storage directly into binary columns—perfect for AI workflows: load PDFs, images, or docs, then parse and analyze with your favorite libraries. Example: read_file('@my_stage/docs/report.pdf') → feed to AI parsers for insights! Dive in: github.com/databendlabs/d…

Databend January 2026 Monthly Report is here! 🚀 - Experimental Table Branching & Tags: Create safe data sandboxes for AI agents with easy DDL, isolation, and snapshot-based time travel. - Spatial Indexes: Full Geometry/Geography support + CREATE/DROP SPATIAL INDEX for blazing-fast queries. - Iceberg Write Support: Upgraded to v0.8.0 for complete write paths and open data lake compatibility. - Query Perf Boosts: Self-join elimination, lazy joins, Parquet optimizations, and high-speed bitmap operations. - Meta Enhancements: Better observability with histograms, faster KV scans, and cluster stability refactors. 11 nightly releases packed with 33 new features and 31 fixes—focusing on governance and speed. Dive in: databend.com/blog/category-…

Excited to explore the future of Agent-Ready Databases—empowering AI agents in enterprise data ops with seamless, secure workflows! Our vision at Databend includes: - Unified semantic storage: Seamless SQL access to structured, unstructured, and vector data—with rich metadata for instant context - Instant schema evolution: Zero-cost changes like adding columns, no data rewrites - Git-like branching: Safe, zero-overhead snapshots for isolated testing and risk-free merges - Robust ACID transactions: Atomic multi-step operations with easy rollbacks for self-correcting agents - UDF Sandbox (coming soon): Isolated Python execution for AI logic, minimizing data risks - Extreme stability: Auto-scaling under high loads for 24/7 agent reliability We're advancing with native Parquet support, metadata versioning, zero-copy snapshots, multi-statement ACID, and more—shaping the next-gen warehouse. Dive in: databend.com/blog/category-…

🚀 We just shipped #Databend MCP: Your safeguard for AI-driven data ops in production. Read-only access meets isolated sandboxes—zero risks, all the power. Key features: • Prod read-only: SELECT/SHOW on live data, no writes • Per-session sandbox: Safe writes in unique namespaces • Verification magic: Sample, simulate, validate queries Integrate with Codex CLI, Claude, Cursor via DSN. Dive in: databend.com/blog/category-… Repo: github.com/databendlabs/m… Docs: databend.com/mcp/

Snowtree GitHub repo: github.com/databendlabs/s…

Introducing Snowtree: Databend's production-proven framework for integrating LLMs into large-scale Rust monorepos. Leverage isolated Git worktrees for safe AI-assisted code generation and refactoring. Execute via raw CLI with models like Codex, Claude Code, and Gemini; enforce rigorous line-by-line reviews to maintain code integrity, mitigate technical debt, and ensure human oversight. Derived from our 1.9M LoC cloud-native data warehouse engineering practices. Explore the details: databend.com/blog/category-…

The architecture:

🚀 Introducing Table Branching in Databend — Git-like version control for your data warehouse tables. Create isolated branches for development, testing, and safe AI agent experiments; tag important snapshots; and query any point in time — all while sharing underlying storage. Here's how it works:

Been waiting to write to Iceberg tables in #Databend? v1.2.867 has you covered: ✓ Full INSERT support ✓ Partition-aware writes ✓ Transaction guarantees ✓ All catalog types supported Check the release notes for details: github.com/databendlabs/d…

Excited to ship branch-aware INSERTs in #Databend! Now you can target specific branches like: ``` INSERT INTO table_name/branch_name ... ``` Big flex for data management. Check it out: github.com/databendlabs/d…

🚀 #Databend Ingestion Benchmarks: Scaling from 3K to 2.3M rows/s! We tested four ingestion methods—Bulk INSERT (with/without presign), STREAMING_LOAD, and STAGE_LOAD (COPY INTO)—in cloud and on-premises environments. The results: Dramatic throughput gains, improved cost efficiency, and insights into the best approach for your workload. Dive into the full analysis: databend.com/blog/category-…