TigerFS

The problem

agents need shared state.
local files weren't built for that.

When agents run on different machines, there's no shared workspace. And even locally, agents can interfere. Coordinating through files means no transactions, no history, and no structure. Git requires pull, push, and merge. S3 has no transactions. Custom coordination APIs have to be built and maintained.

There's a better way.

The ease of a filesystem, the semantics of a database.

Tiger/FS vs. everything else

vs. local files ACID transactions and version history, instead of bare files with no coordination guarantees.
vs. git Changes are visible immediately, with automatic version history. You don't need to pull, push, or merge.
vs. s3 Structured rows and ACID transactions, not blobs. You can query the data, not just retrieve it.
vs. a database Your agents already know how to work with files. You don't need client libraries or schemas to pass around.

Two ways to use it

pick your starting point

File-first

start with files, get a database for free

Write markdown, organize into directories, build lightweight apps on top of the filesystem. Writes are atomic and everything is auto-versioned. Works with Claude Code, Cursor, grep, vim, and anything else that works with files.

See file-first use cases →

Data-first

treat your data as a filesystem

Mount any existing Postgres database and navigate it the same way you navigate directories. Read rows, filter by index, chain queries into paths. The interface your agents already know maps directly onto your data.

See data-first use cases →

Use cases

what you can build

Mode File-first — build apps on the filesystem

shared agent workspace

Multiple agents operate on the same knowledge base at the same time. Every edit is versioned, so if one agent overwrites another's work, you can recover it from .history/.

# agent A writes research findings
cat > /mnt/db/kb/auth-analysis.md << 'EOF'
---
author: agent-a
---
OAuth 2.0 is the recommended approach...
EOF

# agent B reads immediately. no sync. no pull.
cat /mnt/db/kb/auth-analysis.md

multi-agent task queue

todo/, doing/, and done/ are your three directories, and mv is your only API. Moves are atomic database operations, so two agents can't claim the same task.

# agent claims a task atomically
mv /mnt/db/tasks/todo/fix-auth-bug.md \
   /mnt/db/tasks/doing/fix-auth-bug.md

# see what everyone is working on
ls /mnt/db/tasks/doing/
grep "author:" /mnt/db/tasks/doing/*.md

collaborative docs

A human drafts, an agent edits, another summarizes, all in the same directory, all visible immediately. History shows who changed what and when.

# browse the full edit trail
ls /mnt/db/docs/.history/proposal.md/

# see what changed in the last edit
diff /mnt/db/docs/.history/proposal.md/2026-02-25T100000Z \
     /mnt/db/docs/proposal.md

Mode Data-first — treat your data as a filesystem

quick data fixes

Update a customer's email, toggle a feature flag, delete a test record. One shell command instead of opening a SQL client, remembering the schema, and writing a WHERE clause.

# update a single column
echo 'new@example.com' > /mnt/db/users/123/email.txt

# update a full row via JSON
echo '{"email":"a@b.com","name":"A"}' > /mnt/db/users/123.json

# delete a record
rm -r /mnt/db/users/456/

# bulk-load from CSV
cat data.csv > /mnt/db/orders/.import/.append/csv

data analytics

Standard Unix tools work on every table. For larger queries, chain filters and pagination into a single path to push the work into the database.

# find shipped orders
grep "shipped" /mnt/db/orders/*/status.txt

# select specific columns via pipeline
cat /mnt/db/orders/.filter/status/shipped\
    /.columns/id,total,created_at/.export/csv

# sum shipped order totals
cat /mnt/db/orders/.filter/status/shipped\
    /.columns/total/.export/csv | awk '{s+=$1} END {print s}'

up in under 60 seconds.

Works with local and cloud-hosted databases, with special Tiger Cloud and Ghost support.

FUSE on Linux, NFS on macOS, and no external dependencies on either platform.

Install

curl -fsSL https://install.tigerfs.io | sh

Mount and write

# mount a local database
tigerfs mount postgres://localhost/mydb /mnt/db

# or create a new Ghost database
tigerfs create ghost:my-project /mnt/db

# start writing
echo "markdown,history" > /mnt/db/.build/notes
echo "# Hello World" > /mnt/db/notes/hello.md

Read the quickstart

thefilesystemis the API.

agents need shared state.local files weren't built for that.