S-tello Trace · Powered by Stello
Make real work legible to AI.
Find what is worth automating next.
S-tello Trace turns activity across your computer and offline work into verifiable work memory. When workflows repeat, it surfaces system handoffs, human judgment and automation candidates—with evidence and boundaries attached.
- macOS 13+
- Local-first storage
- Screenshots optional
- Invite Qwen3.5 / BYOK / Ollama
How it works
Three steps from scattered activity to actionable context
Start with useful work memory today; move into automation only when repeated evidence is strong enough.
Build work memory
Computer activity, calendars and offline work become a local timeline. Every session can be reviewed, corrected or excluded.
Find repeated workflows
Across multiple days, Trace reconstructs repeated steps, system handoffs and human judgment, then ranks automation candidates by evidence.
Hand off with control
Copy a redacted implementation brief into Claude, Codex or an FDE workflow. Permissions, exceptions and acceptance criteria still require human confirmation.
The morning went to the billing-module refactor: heavy switching between Xcode and GitHub from 9:40 to 11:35, two PRs merged. The 11:40 product sync ran 50 minutes, followed by a Slack backlog sweep. The afternoon's main thread was the quarterly review deck, interrupted three times by support tickets — about 40 minutes total.
Features
Built to answer “what did I actually do today?”
Not another timer — an automatic writer for your working memory.
Narrative daily & weekly reports
Readable paragraphs, not dashboards. Quote them in standups, status updates and reviews.
Calendar-aware
Meetings from your system calendar (and Lark/Feishu) land in the report automatically — work blocks and meeting blocks never blur.
Ready by default, still flexible
Acknowledge the processing disclosure and redeem an invite for hosted Qwen3.5 Flash without running a model locally, or switch to your own API key or local Ollama.
Native menu-bar app
Written in Swift. Light, fast, battery-friendly. Works offline — events persist locally and sync analysis later.
2.1.8 beta hotfix
Your timeline records the apps you actually use again
Fixes an installed-build issue on some Macs that recorded S-tello Trace itself as large event blocks.
Background sampling stays timely
When the app is hidden or covered, foreground app changes continue to be sampled instead of resuming only when Trace is opened.
Trace never records itself
The app is forcibly excluded at event, OCR, retry, and shutdown-write boundaries, so it cannot become a work event.
Existing false blocks are repaired safely
The upgrade quarantines old self-capture and machine-derived blocks while preserving user edits and records backed by real app evidence.
Privacy
First, the straight answer: where your data goes
A tool that records your screen has to earn trust. Here is the S-tello Trace data ledger, line by line.
| Data | Stored | Leaves your machine? |
|---|---|---|
| App & window events | Local SQLite on your Mac | Raw trails are not uploaded; hosted mode sends only bounded text context for the current task through the S-tello AI Gateway, while BYOK connects to your chosen provider |
| Screenshots | Local replay off by default | Not uploaded by default; selected representative frames leave only after separate cloud-visual consent |
| Calendar events | Local cache | Meeting titles join the text summary for report generation |
| Account, invite allowance & submitted feedback | Separate cloud control plane | Email, invite redemption and quota metadata, and minimal feedback fields; sign-in and redemption do not upload work history |
| Crash & health diagnostics | Stored locally first · sending off by default | Preview, dismiss, or consent to one submission; only a separate choice enables future automatic submissions of the same scope. Diagnostics contain no screenshots, OCR, window titles, URLs, raw timeline rows, work content, user input, prompts, or keys. No analytics script on this site |
With no sign-in or remote integrations and a loopback Ollama instance, analysis can stay fully offline. Exclusion rules let you block specific apps and sites (password managers, private browsing). Read the full privacy notes.
For teams & organizations · AI-native
A person sees a day.
An organization sees its work system.
Every process exists in three versions: the one management believes, the one the SOP describes, and the one people actually perform. For AI to take over work it needs the third — and that version lives in no system, only in people's hands, evaporating every evening.
Rank the opportunities
With employees' informed consent, real work evidence shows which steps repeat, across which systems, at what time cost — every automation candidate comes with its evidence and boundaries.
From evidence to agents
Chosen steps grow into editable SOPs and Agent Blueprints, then run as revocable pilots — permissions, approvals and stop conditions stay under human control.
Verify with one ruler
Before and after implementation, the same metrics are compared: hours, volume, error rate, quality. Value counts only when the evidence passes.
How it differs from what you may have considered
| You may have considered | How it usually ends | S-tello Trace |
|---|---|---|
| Employee monitoring | Data polluted by performative work, trust burned — you end up with a leaderboard | Evidence volunteered by employees; you end up with process assets, not a grading tool |
| Consultant interviews | Weeks of interviews, a stack of diagrams that starts aging the day it ships | Real evidence grows into SOPs and Agent Blueprints; the map keeps updating, every claim traceable |
| Buying AI tools for everyone | A little faster, same processes, pilots fizzle out | Automation lands process by process, each verified with before-and-after data |
See clearly first, then build
Stello's forward-deployed engineers (FDE) walk in with Trace: starting from real work evidence, they deliver a ranked automation shortlist, SOPs and Agent Blueprints, then prove the value with controlled pilots.
FAQ
You probably want to ask
Is it watching my screen?
It records foreground apps, windows, browser URLs and timestamps. Page changes and similar boundaries create candidates; only a few key frames survive on-device deduplication. Local replay and cloud visual understanding are both off by default, and cloud visual frames require explicit consent.
Are crashes or hangs uploaded automatically?
No. A diagnostic stays local first; you can preview and dismiss it or consent to one submission. Only a separate choice to send future diagnostics automatically enables later submissions of the same bounded scope. Diagnostics contain no screenshots, OCR, window titles, URLs, raw timeline rows, work content, user input, prompts, or keys.
Will my company use it to monitor me?
No — and it isn't built for that: no scores, no individual rankings. Organization-facing views only contain process data aggregated across multiple people, and raw evidence stays on each person's own machine by default. Team use requires employees' informed consent, and capture can be paused at any time.
Is it free? Will it stay free?
The app is free. Beta users can sign in, acknowledge the managed-AI processing disclosure, and redeem a one-time invite for 5 million tokens of S-tello-hosted Qwen3.5 Flash analysis, valid for 14 days. After that, you can switch to your own API key or local Ollama. Organization-level automation diagnosis and delivery are Stello services, scoped separately.
Do I need my own API key?
No. Fresh installs default to S-tello-hosted Qwen3.5 Flash; sign in, acknowledge the managed-AI processing disclosure, and redeem a beta invite to use it. You can instead bring your own API key or run local Ollama on capable hardware. Raw work trails stay local, and signing in does not upload existing work history.
How is this different from RescueTime, Rize or Timing?
They chart where time went; the output is a graph. S-tello Trace writes what you got done; the output is prose — for standups, updates and honest self-review.
Windows version?
Windows native Early Access is in signing, Store and real-device acceptance. The public experience remains macOS 13+ first for now.
Can I export my data? Am I locked in?
Raw work data is stored in standard SQLite on your own disk and can be exported or deleted. If you use the optional account, signing out or uninstalling does not delete the cloud account, memberships or submitted feedback; cloud deletion is a separate explicit action described in the privacy notes.
Download
See tonight's report about today
Beta 2.1.8, free to download. Work today, read it tonight.
macOS 13 Ventura or later · Developer ID signed and Apple notarized · Release notes & SHA256 · Privacy notes
Invited beta
Keep separate Claude and Codex projects classified separately
Classification Calibration Beta 2.1.8 (219) recognizes distinct session context. Only conflict-free assignments confirmed by the user can affect future records; existing history is not bulk rewritten.
Install once to switch to the isolated Beta update feed; the app then checks every 6 hours and prompts for invited-beta updates · Beta notes & SHA256