npm run lint -- --fix
Executes commands with your permissions — read it before approving.
Keep the workbench. Choose the intelligence. Know who sees every session.
Cloaky is one workbench for your engines and your intelligence — Claude Code, opencode, and Cursor. Start with Venice’s private tier, run local models on your own machine, or connect a provider you already trust — without changing how you plan, build, and review.
How it works PmacOS desktop · public beta · Linux to follow
60+ models · private-tier zero retention
Kimi K3
DeepSeek V4
GLM 5.2
Qwen3from $0.07 / Mtok
Kimi K3DeepSeek V4GLM 5.2Qwen3from $0.07 / Mtok✳ running
Add retry logic to the fetchUser helper and clean up the config.
The helper is called in three places and has no retry handling.
I’ll add exponential backoff with three retries, throw a typed ApiError on final failure, and remove the stale timeout value.
Read · src/lib/api.tsdone
Grep · fetchUser\( · 3 call sitesdone
Edit · src/lib/api.tsupdated
npm test -- --run api6 pass
npm run lint · lastError is unused1 error
The tests pass. One lint issue remains, and a subagent is auditing every caller.
Wrote
src/lib/api.tssrc/lib/api.test.ts✳Thinking…
Tests pass.Fixing the lint error next.
The retry helper is done — how far should the cleanup go?
Should the cleanup carry its own tests?
Local inferenceSearch models…
Sessions on these models run on Ollama, on this machine — prompts never leave it.
Apple silicon · 64 GB unified memory — comfortable to ~38.4 GB per model, tight to 48.0 GB.
Runtimes
Ollama2 modelsServing at http://127.0.0.1:11434. Prompts on these models never leave this machine.
version 0.12.3 · /usr/local/bin/ollama
LM Studionot runningInstalled at /Applications/LM Studio.app, but nothing is answering at http://127.0.0.1:1234.
On this machine 2
qwen3-coderThe most agentic code model of the Qwen family.
30b18.0 GBin picker+1 larger than this Mac
gpt-ossOpenAI’s open-weight models for reasoning and agentic tasks.
20b13.0 GBin picker+1 larger than this Mac
Catalogue
AllCodingFits this MacSelf-improving open models for agentic coding. MIT licensed.
9b5.7 GB35b20.3 GB
deepseek-r1DeepSeek’s first-generation reasoning models.
8b5.2 GB70b42.0 GBtight
qwen3.6Qwen 3.6 — substantial upgrades in agentic coding and thinking.
27b16.4 GB35b20.9 GB
The tests pass. Here is the final review.
The retry helper now retries network failures three times with exponential backoff and throws a typed ApiError. The implementation and its test remain reviewable in the diff pane; the lint issue is still open.
Retry helper · 3 retries + backoffdone
src/lib/api.test.ts6 pass
Lint · unused lastErroropen
Session total: $0.42 · active time: 4m 12s · 18 turns
Add retry logic to the fetchUser helper and clean up the config.
Added the retry helper and updated three callers.
Now chase every fetch wrapper in src/lib.
Add retry logic to the fetchUser helper and clean up the config.
The helper is called in three places and has no retry handling.
I’ll add exponential backoff with three retries, throw a typed ApiError on final failure, and remove the stale timeout value.
Read · src/lib/api.tsdone
Grep · fetchUser\( · 3 call sitesdone
Edit · src/lib/api.tsupdated
npm test -- --run api6 pass
npm run lint · lastError is unused1 error
The tests pass. One lint issue remains, and a subagent is auditing every caller.
Running the API suite with coverage enabled, then reading the report for uncovered retry branches.
Readsrc/lib/api.test.ts
Bashnpm test -- --coverage
GrepwithRetry\( · untested branches
Providers & models
Weaker
Engines
Which agent runs a session. The Claude Code engine is the default and routes through the providers you configure in Providers & privacy; the engines below replace the whole loop with another vendor's agent, under that vendor's account and policies. Pick per thread from the composer.
Claude CodedefaultThe default engine — Cloaky's own loop, routed through the providers you configure in Providers & privacy. Approvals, rewind, MCP toggles, verify-on-done, security scans and thread history are all its features.
Prompts, project files the agent reads, and tool results go to Cursor under your Cursor account and draw on its plan. Cursor states that with Privacy Mode enabled on the account, prompts are not retained or used for training — that setting lives in your Cursor account, applies to the CLI, and Cloaky cannot see or verify it.
Sign-inAuth belongs to the CLI, not to Cloaky — run
cursor-agent login in a terminal once on this machine. Cloaky never stores or proxies that credential.CLI pathUsing /Users/demo/.local/bin/cursor-agent. Set a path only to override.auto-detect
ModelsNot fetched yet — detect to list this account's models in the composer's model picker.
What carries overStreaming, file tracking and the phone companion work. Approval cards appear only when this agent's own configuration asks before an action — OpenCode's defaults ask for nothing, and Cursor gates shell commands but not edits — plus for file writes the agent hands to Cloaky, which follow your permission mode. Rewind, MCP toggles, verify-on-done, security scans and thread history are Claude Code engine features and stay off for these sessions.
Prompts go to whichever provider your OpenCode installation is signed into — that routing is configured in OpenCode itself, and its providers’ retention policies apply. Cloaky cannot see which provider that is.
esc
Cloaky is ready for macOS. Linux is next.
Builds are Developer ID-signed and notarized by Apple — first launch is a double-click. The install guide covers setup.
Add retry logic to fetchUser+24−6
Read · src/lib/api.tsdone
Edit · src/lib/api.tsupdated
npm test -- --run apirunning
Describe the work — / commands, @ files
A complete desktop workbench
Threads, diffs, subagents, readable approvals, and the exact cost of every turn — packaged as a native macOS app.
Bashwrite
npm run lint -- --fix
Runs with your permissions. Review before allowing.
DenyAllow once
Needs input · 1
Readable approvals before execution
File changes and shell commands pause with the exact action in view. Approve once, for the session, or for the project.
Veniceprivate tier · zero retention by contract
Ollamalocal inference · stays on this machineLM Studiolocal inference · stays on this machine
OpenAI · Codexyour account · OpenAI termsGLM 5.2your account · provider termsKimi K3your account · provider termsClaude directyour subscription · Anthropic termsOne workbench, your provider
Start with Venice, run locally, or use a provider you already pay for. Cloaky keeps the active route visible in every session.
v0.1.6public beta · Aug 4, 2026
Local inference becomes a place in the app — browse, fit, pull and run models that never send a prompt anywhere. Read in the changelog →
v0.1.5public beta · Aug 3, 2026
A terminal lives in the workspace now — ⌘J opens your login shell without leaving Cloaky. Read in the changelog →
v0.1.4private beta · Aug 1, 2026
Cloaky now checks for new releases on launch and installs updates with your approval. Read in the changelog →
v0.1.3private beta · Jul 31, 2026
Local model support through Ollama and LM Studio, a local models gallery, hybrid local-plus-Venice sessions, and notarized macOS builds. Read in the changelog →
v0.1.2private beta · Jul 27, 2026
Corrected Venice usage accounting, per-session provider badges, served-model verification, and a live Venice balance. Read in the changelog →
v0.1.0private beta · Jul 25, 2026
First packaged beta: the full Cloaky workbench with Venice inference, privacy-tier model selection, and desktop builds. Read in the changelog →