Install it. Work normally. Export the session to the AI you already use.
local-first work context
Turn work into
AI context
Install → OpenWorkGraph observes how you work locally → export a session to your AI → find repeated work, bottlenecks & automation opportunities
No signup · free to use · data stays local by default · no screen recording · no typed text
Captured work data stays on your computer by default. You decide what leaves.
No screen recording, typed text, clipboard contents or password values in normal capture.
Use it without an account and inspect the code yourself.
example outcome
See what your AI could find.
Pick the kind of work closest to yours and see an illustrative report based on the type of workflow evidence OpenWorkGraph captures.
Closest to your work:
Example AI reportRepeated debugging context gathering
illustrative · one person · one workweek
same context-gathering loop
per occurrence in this example
potential handling time removed / occurrence
What OpenWorkGraph observed
Claude→GitHub→Vercel→Docs→Claude
The real product exports the underlying workflow evidence. Your AI can verify or reject any interpretation instead of relying on a hidden OpenWorkGraph conclusion.
AIPossible workflow improvement
You repeatedly leave the AI to gather deployment state, recent code changes and documentation before continuing the same debugging task.
Give the AI direct access to those three sources through existing connectors, MCP or a small internal context tool. Keep human verification before any production action.
Today
AI → GitHub → logs → docs → AIPossible
AI + connected context → verify → act
Illustrative impact from this example
31 hours / year
If a 6-minute improvement held across 6 occurrences each week for 52 weeks.
$2,340 / year
equivalent capacity at $75/hour
Example only — not a claim about your work or guaranteed savings. OpenWorkGraph provides the real evidence; your AI analyzes your actual workflows.
make the economics yours
What would a small improvement be worth?
You do not need a dramatic automation. Even a few minutes recovered from work you repeat every day can compound. Adjust the assumptions below.
Illustrative capacity recovered
60 hours / year
$4,500 / year
15 min/day × 1 person × 5 days/week × 48 weeks/year.
This calculator does not predict OpenWorkGraph savings. It shows why finding small, real workflow improvements can matter economically. The assumptions are visible and editable.
how it works
Work normally. Make the process visible. Let AI analyze it.
No process-mapping workshop. No need to explain every step from memory. OpenWorkGraph creates evidence from what actually happened.
Use ChatGPT, Claude, browsers, spreadsheets, internal systems and desktop apps as you already do.
09:14 Claude → Browser → Sheets → Claude
OpenWorkGraph records sequence, timing and interaction effort locally — not the text you type.
time · tool · action · effort · context
Export a session to ChatGPT, Claude, Gemini, Grok or another AI, or use experimental MCP access.
workflow.json / .xlsx / .csv → AI
Ask where time disappears, what repeats, what context is missing, and which changes are worth building.
“This lookup repeats 23×. Connect the source directly.”
try it on real work
Install. Work normally. Export.
No account required. OpenWorkGraph runs locally and stores captured data on your computer by default.
Download the .pkg, double-click it, then open OpenWorkGraph from Applications. No Terminal or Python setup.
ZIP is the manual/tester bundle. For the easiest setup, use the installer above. Download links always follow the latest GitHub release.
Early prototype: the installer may show an unverified-developer warning until platform signing is enabled.
Advanced / manual install +
For technical users or troubleshooting only. The normal path is the installer above.
Already connecting through ChatGPT? Setup guide →
Free · open source · stored locally by default · no screen recording or typed text · view source ↗
privacy boundary
Useful context without turning your computer into a surveillance feed.
OpenWorkGraph is designed to capture the path and effort of work while deliberately excluding the most invasive forms of observation.
It sees
- Apps and windows
- Browser tabs and navigation
- Time and activity
- Clicks and scrolling
- How much typing happened
- Some safe page and control labels
It doesn’t record
- What you typed
- Which keys you pressed
- Clipboard contents
- Password-field values
- Selected text
- Your screen or microphone in normal operation
Rich exports can still contain useful business context such as window/page titles, company or project names, amounts, and safe UI labels. You can inspect the export before sharing it with an AI or anyone else.
what works today
Capture now. Bring your own AI.
The core observer, browser context and local exports work today. OpenWorkGraph does not require a built-in model: export the evidence to the AI you already trust and use.
Desktop observer
Apps, windows, timing, clicks, scrolling and aggregate keypress effort.
AvailableBrowser context
Tab navigation and semantic browser activity.
AvailableLocal export
Export a session to the AI tool you already use so it can analyze how the work happened.
AvailableMCP / agent access
Authenticated direct context access for supported AI tools and MCP clients.
ExperimentalManaged team deployment
Organization-wide rollout and administration.
In progressquestions
Small print, in plain English.
What does OpenWorkGraph actually do?+
It records how work moves across apps and browser tools — sequence, timing and interaction effort — and turns that into structured evidence you can give to AI to investigate how the process could be faster, simpler or more automated.
Is the example AI report real?+
It is an illustrative example of the kind of analysis a model could perform on OpenWorkGraph evidence. The numbers are deliberately labelled as examples, not measured savings or promises about your work.
Why can’t I just tell ChatGPT or Claude how I work?+
You can, but people rarely remember every tool switch, repeated lookup, delay and interaction accurately. OpenWorkGraph creates evidence from what actually happened, so the AI can reason from the real process rather than a summary from memory.
Does it record what I type?+
No. It can count how much typing happened. It does not record which keys, the order of keys, typed field values, or clipboard contents.
Where does my data go?+
Captured data is stored on your computer by default. It only leaves when you deliberately export or share it, or connect OpenWorkGraph to another system.
Is it free?+
Yes. OpenWorkGraph is free and open source. You can inspect the code and the data it produces yourself.
Is this employee scoring software?+
No. It is designed to reconstruct work processes and context, not rank people.
Can I use it with AI now?+
Yes. The simplest route is to export a session and upload it to ChatGPT, Claude, Gemini, Grok, or another AI that accepts files. OpenWorkGraph also includes experimental MCP access for supported clients.
find the opportunities in your own work
Make the work visible.
Let AI improve it.
Install OpenWorkGraph, capture a real session, and give your AI evidence of where time, effort and money may be disappearing.
Free · open source · stored locally by default · no screen recording or typed text