Run Kimi Code from anywhere

Every Kimi Code session gets its own git worktree, its own Postgres, its own ports. Parallel agents never fight over your dev database — and you keep coding in yours.

View on GitHubRead the docsInstall CLI
All install options
Works with
Kimi Code
thinslice terminal showing slice tmp --from current kimi running in an isolated worktree

Keep coding in your worktree
while Claude works in its own.

No shared database, no port collisions — isolated by default.

thinslice slice list output
thinslice Kimi Code session in isolated worktree
thinslice state diff view

Every slice has its own everything

thinslice binds each git worktree to its own bag of state — no shared databases, no shared ports, no shared config. Kimi Code runs inside a slice and can't reach yours.

Postgres
Redis
Ports
Config
Git worktree
Read the docs

One command to launch

Spin up a fresh worktree with cloned state and hand it to Kimi Code in one shot. thinslice handles the postgres clone, the redis snapshot, and the port reservation — you just describe the task.

slice tmp
--from current
kimi "…"

Isolated worktree

with

Cloned state

Own Postgres
Own Redis
Own ports

Review, preview, ship

When Kimi Code is done, review the diff inline, run your app against the isolated database, then commit and open a PR — without ever touching your main working tree.

fix-authworktree
WorktreePreviewReviewCommitPRMerge
web.fix-auth.my-app.localhost
Inline review3 files changed
src/auth/session.ts+42
src/auth/middleware.ts+18-9
tests/auth.test.ts+31
Ready to shipchecks passed
Commit
Open PR
Merge

Forget about ports

When Kimi Code sessions run in parallel, each slice gets its own reserved port range. No conflicts, no guessing which process is on which port.

my-app
fix-authnpm run dev
:3001
add-searchnpm run dev
:3002
upgrade-depsnpm run dev
:3003

Fully scriptable

Everything you can do interactively, you can do from a script or CI pipeline.

slice tmp --from current -- claude -p "implement user authentication"slice tmp --from current --name feature-x -- claude -p "implement feature X"slice ls                                          # list running slicesslice status feature-x                        # check a sliceslice rm feature-x                            # tear down + clean state
Full CLI reference

FAQ

+Do I need a Moonshot AI API key?
Yes. Kimi Code CLI requires a Moonshot AI API key to call the Kimi models. thinslice doesn't add any credential requirements — it just launches Kimi Code in an isolated environment using whatever credentials Kimi Code already has configured.
+Is this free?
Yes. thinslice is free and open source. You need Kimi Code installed separately with your own credentials. thinslice only manages the local worktree and state — your Kimi Code usage goes through Kimi Code as normal.
+Does my code or data leave my machine?
No. thinslice runs entirely locally. It clones your Postgres database, binds a port range to the slice, and runs Kimi Code as a subprocess. Nothing is sent anywhere by thinslice.
+What does "state" mean exactly?
Each thinslice slice gets its own Postgres instance (cloned from yours), its own Redis instance, an isolated set of reserved ports, and environment config scoped to that worktree. Kimi Code runs inside this environment and can't accidentally mutate your main database.
+Do I need Docker or Kubernetes?
No. thinslice uses native OS process isolation and Postgres's built-in copy capabilities. No containers, no VMs, no root access required.
+What happens when I'm done with a session?
Run 'slice rm <name>' and thinslice tears everything down — the worktree, the Postgres clone, the Redis instance, the reserved ports. Your main environment is untouched.
+Can I run multiple Kimi Code sessions at once?
Yes, that's the main use case. Each 'slice tmp' call creates a fully independent environment. You can run as many parallel Kimi Code sessions as your hardware supports, each with its own database and ports.
+Do I need git?
Yes, git worktrees are how thinslice isolates the code. You don't need GitHub or a remote — a local repo is enough. thinslice creates the worktree branch automatically.
+How does the Postgres clone work?
thinslice uses Postgres's CREATE DATABASE … TEMPLATE to snapshot your database in place. On macOS with APFS, the data files are copy-on-write so the clone is near-instant regardless of database size. See /docs/ for details.

thinslice gives every coding agent its own isolated slice of your stack. No Docker, no VMs, no shared databases. Agents work in parallel without stepping on each other.

If thinslice has been useful and you want to say thanks, or if you just want to talk shop about agents, evals, sandboxes, or go-to-market for automation products, give me a call.

Book 30 minutes