SyneThe Context Compiler.

Compile a Java or PL/SQL repository into a queryable code graph, then hand an AI agent only the slice it needs. Runs entirely on your machine.

Get Started

View on GitHub

A queryable code graph

Parses a git ref into typed nodes — files, types, callables, SQL statements, tables — joined by structural, reference, call-flow and data edges.

Three questions, answered

Blast radius (what breaks if this changes), blueprint (structure without the bodies), and call stack (entry point down to the tables it touches).

Wired into your agent

A read-only MCP server serves the graph to an LLM or IDE over stdio, and one wire command registers it with your agent harness.

Status

Early prototype, no public release yet. The command surface and the graph schema will change.

How it works

bash

# 1. Compile the repository at a git ref into a code graph
$ syne index . -b main --stats

# 2. Find every reader and writer of a table
$ syne blast-radius table:PAYMENTS

# 3. Trace an entry point down to the data it touches
$ syne call-stack 'myapp::Callable:src/main/java/com/bank/PaymentProcessor.java#process()'

# 4. Serve the graph to an agent (read-only, stdio)
$ syne mcp

Why compile context at all?

Dumping 50,000 lines of Java into a context window costs tokens, buries the code that matters, and invites hallucination. Syne stores metadata only — signatures, spans, and a confidence score on every edge — never method bodies, which it hydrates on demand. Unresolved calls are reported as unresolved, so a missing edge means not seen, never does not exist.

That matters most on the codebases nobody wants to touch: large legacy Java and PL/SQL estates where the blast radius of a change is unknown. Syne maps it locally, before a line is generated.