alpha · claude code
your claude code sessions, searchable.
you've built more than you remember.
pip install getflex
In Action
trace every file. every decision.
Every session you've had with Claude Code, compiled into one SQLite file with structured retrieval. Run flex init then ask anything — no API, no cloud, no setup.
run demo
Run Again
five calls. one answer.
zero indirection.

Claude ran @orient to read the schema, then assembled the full answer in four targeted queries.

  • 01 @orient — schema, views, presets, graph intelligence. Claude knows what's available before writing a single query.
  • 02 _edges_file_identity — stable UUID across every path the file has ever lived at.
  • 03 messages — every session that touched it, in order.
  • 04 vec_ops(...) — semantic context: why it was built, what it replaced.
  • 05 session content — the exact prompt that spawned the file.
3,337
sessions
148K
chunks
1
SQLite file
Capabilities
your codebase has photographic memory.
Every session compiled into structured data. Graph intelligence added automatically. Hooks keep it current — updates on every new message.
Semantic Search

"Why did we switch to Postgres?" Claude finds the session where you decided. Ask anything — it searches across every project, every session.

Knowledge Graph

Claude knows which sessions and files matter most. Centrality, communities, hubs — pre-computed so every query is already weighted by importance.

File Lineage

Ask about any file. Claude traces it through every session that ever touched it — who created it, why, what changed, what it became.

Your Projects
every repo. every conversation. one interface.
Sessions grouped by project automatically. Search one repo or all of them at once.
why did we drop the redis cache layer? all projects
project sessions chunks status
core-api 533 sessions 15K chunks indexed
agent-runtime 158 sessions 15K chunks indexed
search-engine 90 sessions 4.2K chunks indexed
+ yours
What Gets Compiled
nothing is thrown away.
Other tools extract facts. Flex keeps everything — and lets Claude decide what's relevant at query time.
All Messages
Every prompt, every response — the full conversation, not just the outcome.
Tool Operations
Every file touch logged — Edit, Write, Bash — with the session that ran it.
Sub-Agents
Every delegation tracked — which agent ran what, in what order, with what result.
File Tracking
Stable UUID. Tracks the file through renames, moves, repo migrations.
Soft Provenance
Every Bash command captured — moves, deletions, side effects inferred from context.
Full Fidelity
Not extracted entities. Not a summary. The actual conversation.
Local-first
own your data.
Your entire knowledge base is one SQLite file on your machine. ls it. Ship it. Back it up. Run it offline. No cloud. No API key. No vendor. No rate limits.
# your entire knowledge base is one file
$ ls ~/.flex/cells/
claude_code.db

# back it up
$ rsync -av ~/.flex/cells/ backup:~/
claude_code.db    284M

# ship it
$ scp claude_code.db prod:~/
claude_code.db    100%   284MB/s

# open format — query it directly
$ sqlite3 claude_code.db "SELECT COUNT(*) FROM sessions"
3337