The question is in English. The data isn't.

Jacynth translates plain-language questions into SQL, Cypher, and Elasticsearch queries — checks them for safety, runs them, and streams the answers back live. Inference happens locally through Ollama, so the question never leaves your infrastructure.

View on GitHub

Show me BTC volume spikes in the last hour

// target: materialize — validated, read-only
SELECT window_start, sum(volume) AS vol
FROM crypto_trades WHERE symbol = 'BTC'
AND ts > now() - INTERVAL '1 hour'
GROUP BY 1 HAVING sum(volume) > 2 * avg_baseline()
// → streaming 14 rows over WebSocket

Why it holds up

Translation is easy. Trustworthy translation isn't.

Safety firstEvery generated query is validated before it touches a database: injection prevention, permission checks, and read-only enforcement for anything untrusted.
Local inferenceOllama runs Llama 3.1, Mistral, or your fine-tuned model on your own hardware. Private by architecture, not by promise.
Live resultsContinuous queries on Materialize stream over WebSocket — dashboards and alerts update themselves.
It learns your schemaThumbs up or down on results feeds back into query generation. Your team's domain knowledge accumulates instead of evaporating.
API-firstOne REST/WebSocket call drops Jacynth into Versal, Akashic, Delt, or your own tools.

Query targets

Four backends, one interface

Jacynth picks the right query language for wherever your data actually lives.

streaming SQL

Materialize

Live dashboards, continuous queries, event-driven results.

SQL

PostgreSQL

Joins, aggregations, and analytics on relational data.

Cypher

Neo4j

Relationship traversal and knowledge-graph exploration.

DSL

Elasticsearch

Full-text search, log analysis, and large aggregations.

Where it runs

Already answering questions across the stack

Versal uses it for natural-language market queries in the terminal. Akashic routes code search through it — find functions, trace dependencies, in English. Delt asks it for portfolio analysis and transaction history. Apollo orchestrates the model side, routing each Jacynth request through the optimal LLM.

Pricing

One small line item

$6/mo

as an add-on to any Colossal Capital tier. The tier (free and up) sets query capacity, API limits, and team size. Sign up at the CC keys portal · tier details