Streamlit st.session_state limits: server-side memory and per-tab session scope

Direct answer

There is no documented hard size limit for st.session_state. The practical ceiling is the RAM available to the Streamlit server process. Objects remain in memory, uncompressed, for the lifetime of the session, so a DataFrame costs approximately its in-memory Python size.

Per-tab isolation means each browser tab receives its own session and its own copy of st.session_state. One user with five tabs holding the same 100 MB DataFrame can consume roughly 500 MB for that object alone, plus interpreter overhead. Aggregate memory scales approximately linearly with concurrent tabs.

Keep user-specific, transient data in st.session_state: widget values, form inputs, current step, short-lived intermediate results. Use st.cache_data for expensive, shareable, read-only computations reused across sessions. Move large, long-lived, or multi-user datasets to an external database or object store.

Confirmed facts and likely explanations

Item	Status	Detail
Hard st.session_state size limit	Not documented	Bounded by server process RAM; no automatic compression or paging.
Per-tab memory scaling	Expected behavior	Independent sessions; memory roughly proportional to tab count for identical objects.
Session state lifetime	Documented	Persists across reruns; released when the session ends.
st.cache_data sharing	Documented	Can share computed results across sessions, but still consumes RAM until evicted.

Verification steps

1. Run a minimal app that assigns a generated DataFrame of known target size, such as 100 MB, to st.session_state.df.
2. Measure the Streamlit process resident memory before and after assignment using psutil or docker stats.
3. Open the app in multiple tabs and confirm memory increases approximately in proportion to tab count.
4. Repeat with st.cache_data and compare process RAM; cached data may be shared but is not automatically evicted on session end.

import os
import psutil
process = psutil.Process(os.getpid())
print(f"{process.memory_info().rss / 1e6:.1f} MB")

Missing diagnostic detail

Confirm the expected peak number of concurrent sessions or tabs on the server. That figure, multiplied by the per-session object size, determines whether session state fits in available RAM or whether external storage is required.

Assumptions and uncertainty

This draft assumes Streamlit 1.x server-side execution. Memory figures are illustrative and require verification against your app, data size, and server environment. Because current verification has not been completed, treat specific size thresholds as provisional.