// the find
AltimateAI/altimate-code
Open-source agentic data engineering harness for dbt, SQL, and cloud warehouses. 100+ tools, 10 warehouses, AI-powered.
A terminal-based AI coding agent forked from OpenCode, specialized for data engineering teams. It wraps an LLM with 100+ deterministic tools for SQL analysis, column-level lineage, dbt workflows, and cross-warehouse operations — acting as the data-aware tool layer that general coding agents lack. Target audience is data/analytics engineers who work daily with dbt, Snowflake, BigQuery, or Databricks.
The deterministic SQL analysis is the real differentiator — 19 anti-pattern rules with published benchmarks (1,077 queries, zero false positives) beats LLM pattern-matching for correctness on known problems. Column-level lineage extraction across dialects is genuinely hard to get right and they've benchmarked it properly. The three-mode agent design (builder/analyst/plan) with SQL write-access gating is a sensible safety model for production warehouse connections. Cross-dialect data parity via joindiff/hashdiff without data egress is a real operational need that most teams solve with janky one-off scripts.
It's a fork of OpenCode — the core agent loop, TUI, and streaming code isn't theirs, so the moat is only as deep as the data tools on top. The benchmark methodology is self-reported with no independent replication, and '100% F1' on any real-world benchmark should trigger skepticism. The 'AI Teammate Training' feature that 'learns from examples and applies rules automatically' is described vaguely with no explanation of how rules are stored or enforced — could be prompt injection waiting to happen. Windows on ARM64 is unsupported via the standalone binary, which matters more as ARM developer machines grow.