SynapCores v1.14.0 — Your database can see now

For years the deal with "AI vision" was the same: ship your image to someone else's server, pay per call, wait on their rate limits, and hope your data was allowed to leave the building. Scanned invoices, product photos, inspection shots of a failing part on a line — all of it had to become someone else's traffic before you got a single sentence back.

Not anymore. v1.14.0 gives SynapCores native, in-process vision — it looks at an image and tells you what it sees, from inside your own database process.

One endpoint. One question. A real answer.

curl -X POST http://127.0.0.1:8080/v1/multimodal/describe \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{ "prompt": "What is wrong in this photo?", "image": "<base64>" }'
# → { "text": "A hairline crack runs along the weld seam of the lower bracket…",
#     "model": "llava" }

That's it. The image is decoded and interpreted by a vision model running in the same process as your tables — no daemon, no separate service, no round-trip.

Why it matters

Private by default. The image never leaves the machine. No egress, no third-party logs.
No cloud, no keys. The vision model runs in-process. If SynapCores is running, vision is running.
Runs on plain CPUs. Validated on commodity, non-AVX-512 hardware. No GPU fleet required.
Bring your own model — or ours. Defaults to LLaVA; point it at the vision model you prefer per request. Cloud providers (OpenAI, Anthropic, Ollama) still work when you want them — you just no longer need them.

Pull a vision model into the in-process registry once, and it's ready:

docker run --rm -v synapcores-data:/var/lib/synapcores \
  synapcores/community:v1.14.0-ce pull llava

Give an in-database AGENT_RUN workflow eyes; read scanned documents the moment they land; triage inspection photos next to the rows that describe the asset. The gap between "data" and "understanding" keeps shrinking — this release deletes another piece of it.

Validated

feature_validator 88/88 · recipe certification 158/162 with zero regressions vs the prior release · non-AVX-512 canary boots and runs the models with no illegal-instruction · native vision end-to-end described a test image in-process on CPU, on both the raw binary and the published Docker image.

Get it

docker pull synapcores/community:latest
# or a pinned version:
docker pull synapcores/community:v1.14.0-ce

Drop-in upgrade — no schema migration. The vision route is additive; every existing path is unchanged.

Your data already knows more than it's telling you. Now it can show you.