jammi-client 0.26.4

Pure-Python gRPC client for a remote Jammi engine — the deploy half of jammi-ai

jammi-client

Pure-Python gRPC client for a remote Jammi engine — the deploy half of the develop→deploy journey.

jammi-ai runs an embedded engine in-process and bundles this client for its remote targets. jammi-client is the lean variant: a universal py3-none-any wheel that links no candle/ML stack — just grpcio + protobuf

• pyarrow — for production where the engine runs behind a server.

One front door

import jammi_client

db = jammi_client.connect("https://engine.example.com")
db.add_source("patents", url="s3://corpus/patents.parquet", format="parquet")
db.generate_embeddings(source="patents", model="local:tiny_bert",
                       columns=["abstract"], key="id", modality="text")
q = db.encode_query(model="local:tiny_bert", query="quantum computing")
hits = db.search("patents", query=q, k=5)   # -> pyarrow.Table

connect(target) is the Python mirror of the Rust Jammi::open(Target): transport is config, not a code path.

Authenticated channels

A bearer-protected endpoint is reached by attaching credentials to the channel — the bearer rides the channel (attached once at connect), not threaded through every call:

from jammi_client import connect, BearerCredentials

db = connect("https://engine.example.com", credentials=BearerCredentials(token))

The same works on a plaintext grpc:// target for local development. The channel-level bearer covers the typed gRPC verbs; db.sql() (the Flight SQL lane) does not yet carry it — tracked at issue #96.

target	transport
https://host / grpcs://host:8081	secure remote
http://host / grpc://host:8081	plaintext remote
file:///data	local engine — not in this build; raises NoEmbeddedEngineError pointing at pip install jammi-ai

Scaling local→remote is an env flip (connect(os.environ["JAMMI_TARGET"])) with no code change. Productionising from the embed wheel to this lean client is a one-line import swap (import jammi_ai → import jammi_client), connect unchanged.

Generated from the canonical proto

The wire stubs under jammi_client/_generated/ are generated from the same jammi.v1 proto the engine, npm client, and embed wheel speak — one source, no parallel schema. Regenerate after a proto change:

pip install -e '.[dev]'
make generate