SSpine3.33.0
the full feature set · verified at 3.33.0

Every capability, and nothing we can't evidence.

Package synaptixs-spine · CLI orchestrator · MCP server orchestrator-mcp

MIT licensed · deterministic, no-LLM comprehension · every fact carries file:line provenance

9languages
9doc formats
55CLI commands
32MCP tools
8 / 11node / edge kinds
1.00graph precision

Browse the full capability matrix — by category, with status →

Languages parsed · 9

Comprehension is 9 front-ends, not 40 — with precision 1.00 on every node and edge kind, measured against a published corpus covering 8 of them (PHP, the newest, has no accuracy numbers yet).

PythonJavaTypeScriptC#CC++GoPHP SQL data layer

Document formats · 9

Each becomes a Doc node. The deterministic build is text-only; image OCR and audio/video transcription are opt-in extras.

.md.markdown.rst.txt.html.htm.pdf.docx.xlsx

.pdf / .docx / .xlsx need the [docs,office] extras

Comprehension · 8 read-only tools

Free wedge, no credentials →

Writes nothing. Ask in plain language — the assistant picks the tool.

map_repoblast_radiusexplain_symbolinvestigatelocalizeregression_gapsroot_causedocs_for

+ gated SDLC tools complete the 32 total — those write, and only after a human gate.

CLI · 55 commands, 7 areas

Driven by orchestrator.

  • understand extract & query the graph
  • blast_radius what breaks if I change this
  • regression reachable, untested symbols
  • sdlc grounded codegen + tests + PR
  • doc_drift docs lying about the code
  • audit replay the append-only log

Governance

Buyer page →
  • ▸ 🔒 Two human gates — intent, then PR approval
  • ▸ Policy-as-code, reviewed in version control
  • ▸ Append-only audit — nothing rewritten
  • ▸ Replayable runs · durable execution (Temporal)
  • ▸ Budget caps on any autonomous run
  • ▸ Safe by default — --safe writes nothing external

Optional extras

Install only what you need: pip install 'synaptixs-spine[docs,office]'

alllanguagesdocsofficemcpsqlgophpmediaasrotel+ more
what Spine deliberately does not do
✕ No 40-language claim — comprehension is 9 front-ends
✕ No LLM in the deterministic build — media (OCR/ASR) is opt-in, off by default
✕ No push, merge or tracker write without approval
✕ No external write under --safe
✕ No deployment — the pipeline ends at the reviewed PR
✕ No unmeasured claims — accuracy is precision/recall on a published corpus

Admitting boundaries is exactly what makes the rest credible.