Describe the stack that runs, not the one that is planned

[DenisDrobyshev]

Two claims in the stack paragraph name components that are not there.

Postgres with pgvector

grep -rn "psycopg" --include=*.py .          -> no matches
grep -rln "postgres|POSTGRES|DSN|dbname" src -> no matches
find src/praxis/index -type f                -> schema.sql only

docker-compose.yml starts a db service the app service does not depends_on and never connects to. What serves dense retrieval, per its own docstring in retrieve/dense.py, is a matrix-vector product over vectors held in memory and cached to disk, with a pure standard-library fallback — which is exactly what the offline image uses.

GigaChat and YandexGPT

ls src/praxis/llm/  ->  __init__.py  anthropic_client.py  base.py  mock.py
grep -rniE "gigachat|yandex" --include=*.py src/  -> no matches

Claude and a deterministic mock are implemented. Neither Russian provider is, and no environment variable for either exists in the source — ANTHROPIC_API_KEY is the only one.

This is the more consequential of the two. Data residency is a compliance argument. Someone evaluating praxis for a Russian client reads "RU-провайдеры (GigaChat, YandexGPT) для сценариев с требованием резидентности" as a capability they can switch on, and it is a plan.

What the paragraph says now

Both are described as designed or next, and each names the file that shows the current state: src/praxis/index/schema.sql for the Postgres path, src/praxis/llm/base.py for the provider contract. A reader can check either claim in one click, which is the property the whole section was missing.

The website carried the Postgres claim in a stronger form — it explained why keeping the index in the same database was a good design — and was corrected in DrobyshevDev.github.io#14. This is the source of that claim.

Context and the three options for actually closing the Postgres gap: #26.