Altimate Code – Open-Source Agentic Data Engineering Harness
I'm Anand, co-founder and CTO of Altimate AI. My co-founder Pradnesh and I are open-sourcing Altimate Code. AMA. Why we built this: Pradnesh and I have been building tooling for data engineers for three years: dbt Power User and Datamates vscode extensions with combined 750k+ installs, running against real Fortune 500 data estates. The pattern we kept seeing: general-purpose agents can write SQL, but they have no model of what the SQL does. No lineage. No schema context. No understanding of what's in a dbt manifest. That's not a prompt problem; it's a missing tool layer problem. The numbers…
In plain words
Altimate Code is an open-source agentic data engineering tool designed for data engineers working with dbt projects. It addresses limitations in general-purpose AI agents by providing SQL generation with understanding of data lineage, schema context, and dbt manifests. The tool aims to reduce common errors in AI-generated SQL, such as references to non-existent tables and silent join failures that produce incorrect results without warnings.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
I'm Anand, co-founder and CTO of Altimate AI. My co-founder Pradnesh and I are open-sourcing Altimate Code. AMA. Why we built this: Pradnesh and I have been building tooling for data engineers for three years: dbt Power User and Datamates vscode extensions with combined 750k+ installs, running against real Fortune 500 data estates. The pattern we kept seeing: general-purpose agents can write SQL, but they have no model of what the SQL does. No lineage. No schema context. No understanding of what's in a dbt manifest. That's not a prompt problem; it's a missing tool layer problem. The numbers make it concrete: 27–33% of AI-generated SQL references tables that don't exist. 78% of errors are silent wrong joins, queries that compile, run, and return confidently incorrect data. One team got a $5k bill from a single Cortex AI query their resource monitors never caught. This isn't a model quality problem. It's a missing harness problem, and we proved it. Claude Code and Cursor are genuinely good for software engineering. But when you point them at a data stack, they hallucinate column names, ignore partition keys, and have no concept of data contracts or quality rules in your models. From building tooling against real data estates, we knew exactly what was missing at the tool level. We forked from OpenCode for the agentic scaffolding. What we added is the entire data layer: compiled Rust engines, purpose-built skills, and the harness that wires them together. What Altimate Code does that general agents can't: - Live column-level lineage: traces any column through joins, CTEs, and subqueries deterministically. 100% edge match on 500K benchmark queries at 0.26ms/query, and not from a cached manifest. Manifests go stale within hours on active pipelines, which makes cached lineage unreliable for anything agentic - SQL anti-pattern detection: 26 rules, zero false positives, 0.48ms/query. - Local SQL validation: interrogates your schema catalog in 2ms without touching your warehouse. Wrong table? Caught with a fuzzy-matched fix suggestion before the LLM goes into a fix loop. That's 10ms for 5 fix cycles vs. 2.5 minutes of Snowflake round-trips - Purpose-built skills for dbt development, testing, troubleshooting, documentation, SQL optimization, and migration. - 3 agent modes with compiled permission enforcement (Builder, Analyst, Planner). “Analyst” enforces read-only at the engine level, not just the prompt. That distinction is what makes it safe to run against production - Persistent memory: cross-session, two scopes (global preferences + project knowledge). Versioned in git, team-inherited on git pull - PII detection, SQL injection scanning, permission enforcement — all at the engine level, not the prompt - 10 data connectors: Snowflake, BigQuery, Databricks, PostgreSQL, Redshift, DuckDB, MySQL, SQL Server etc - Local tracer: every LLM call, tool invocation, and warehouse credit traced locally. No external services. On the benchmarks: We ran ADE-bench, the open standard from dbt Labs Altimate Code (Sonnet 4.6) → 74.4% Cortex Code (Snowflake) (Opus 4.6) → 65% Claude Code (baseline) (Sonnet 4.6) → ~40% A cheaper model with compiled tools outperformed a more expensive model without them. The gap is the harness. Full methodology is in the launch post and linked from the README. To try it: - npm install -g @altimateai/altimate-code - altimate - altimate /discover /discover interrogates your dbt projects, warehouse connections, and installed tools automatically. GitHub: https://github.com/AltimateAI/altimate-code · Docs: [altimate-code.sh](http://altimate-code.sh) There's a /feedback command that files a GitHub issue directly. If something breaks or doesn't behave the way you'd expect, use that or reply here. I'll be in this thread.
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 16d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, March 2026
the whole month →
Switch from ChatGPT to Claude with import memory feature
AI · Mar 2026 · claude.com
