Skip to content

Self-improving AI Employees for everyone

docs: add AI employee work-cycle explainer

trohitg7611ef3a6519
diff --git a/README.md b/README.md
index 2cac34a4..11d20d3a 100644
--- a/README.md
+++ b/README.md
@@ -56,9 +56,11 @@ The builds are not code-signed yet. If your computer blocks the installer, macOS
### How an employee works
-[![Connect AI, describe a job, review the setup, hire, and coach your employee](docs/diagrams/how-it-works.svg)](docs/diagrams/how-it-works.svg)
+[![How AI employees work: jobs arrive through chat, email, messaging, schedules or webhooks; an employee team uses browsers, apps and computer tools to deliver results, with memory, skills and feedback informing the next job](docs/diagrams/how-ai-employees-work.png)](docs/diagrams/how-ai-employees-work.png)
-An employee takes a job through the steps needed to complete it: browsing websites, using connected apps, working with files, running code, and coordinating specialists. It can work when a message arrives, at a scheduled time, or when you ask it to. Its lead assigns work and reviews submissions. You can see progress, results, and requests for help on the employee's page.
+Jobs can arrive through Home chat, email (Gmail, Outlook, or IMAP), WhatsApp, Telegram, or Discord, or start on a schedule. Dev mode also supports webhook inputs.
+
+An employee takes a job through the steps needed to complete it: browsing websites, using connected apps, working with files, running code, and coordinating specialists. Its lead assigns work and reviews submissions. You can see progress, results, and requests for help on the employee's page.
**You set the ground rules.** New hires ask before sending by default. Supported send actions appear as drafts for approval; browser use is restricted to reading while Ask first is on. You can review app-access requests, change access, and stop or resume the employee from Home. Keep the app running and your computer awake for background work.
diff --git a/docs/diagrams/README.md b/docs/diagrams/README.md
index 4e687514..e98c64d9 100644
--- a/docs/diagrams/README.md
+++ b/docs/diagrams/README.md
@@ -1,11 +1,20 @@
# Diagrams
-One set, dark theme, all SVG. This folder is the only copy; nothing under
-`docs-internal/` vendors these files.
+Dark-theme architecture diagrams in SVG and a generated product explainer in PNG.
+This folder holds the project assets; nothing under `docs-internal/` vendors them.
**Overview diagrams** (linked from the README and CONTRIBUTING):
-- `how-it-works.svg` (README's guide for everyone: connect AI, describe a job,
+- [how-ai-employees-work.png](how-ai-employees-work.png) (README's guide for
+ everyone: input channels, employee team, browser/app/computer work, results,
+ and the memory/skills feedback loop). Inputs show Chat, Email, WhatsApp,
+ Telegram, Discord, Schedules, and Webhooks (Dev mode).
+ App tiles show Gmail, Google Calendar, Google Drive, Google Sheets, GitHub,
+ and Stripe.
+ Created with built-in ImageGen using the
+ onboarding's product style; the [generation prompt](how-ai-employees-work.prompt.txt)
+ records the brief and source references. It is a conceptual product illustration.
+- `how-it-works.svg` (editable hiring overview: connect AI, describe a job,
review the setup, hire, and coach)
- `default-workflows.svg` (README's technical section: example workflow structures)
- `system-overview.svg`, `execution-flow.svg`, `ai-agent-routing.svg`, `node-anatomy.svg` (CONTRIBUTING)
diff --git a/docs/diagrams/how-ai-employees-work.png b/docs/diagrams/how-ai-employees-work.png
new file mode 100644
index 00000000..b851dd48
Binary files /dev/null and b/docs/diagrams/how-ai-employees-work.png differ
diff --git a/docs/diagrams/how-ai-employees-work.prompt.txt b/docs/diagrams/how-ai-employees-work.prompt.txt
new file mode 100644
index 00000000..e03bd13c
--- /dev/null
+++ b/docs/diagrams/how-ai-employees-work.prompt.txt
@@ -0,0 +1,72 @@
+OpenCompany AI employee work-cycle image
+Generated: 2026-10-24
+Mode: built-in ImageGen (generation and targeted edits)
+Asset: docs/diagrams/how-ai-employees-work.png
+Purpose: README / How an employee works
+User choices: polished static image; onboarding product visuals; full work cycle.
+Sources: docs-internal/onboarding.md; design_handoff_onboarding/README.md; client/src/features/home/onboarding/steps/WelcomeDemo.tsx; docs/diagrams/how-it-works.svg; server/config/employee_apps.json; server/services/employees/builder.py; server/nodes/trigger/webhook_trigger/__init__.py.
+Reference images sent to ImageGen: none for the initial generation. The targeted edit used docs/diagrams/how-ai-employees-work.png as its edit target.
+User refinements: add the proper input channels in the Give a job section; add recognizable supported apps to the Apps & integrations UI.
+Integration sources: server/config/employee_apps.json; server/nodes/github/__init__.py.
+
+Initial generation prompt:
+
+Use case: infographic-diagram.
+Asset type: a polished static explanatory image for the OpenCompany GitHub README, for nontechnical readers.
+Primary request: Explain how AI employees complete work inside OpenCompany, an open-source operating system for AI employees. Show the FULL work cycle: a person gives a job, an AI employee lead coordinates specialists, the team uses browsers, apps, integrations and the computer, completed work appears in Workspace, and feedback plus memory and skills inform future work.
+
+This is a new image. Use the visual language of the project's onboarding: calm, high-fidelity desktop-product graphics, rounded cards, hairline borders, restrained floating shadows, pill status indicators, circular initial avatars, simple line icons, and ample negative space. Create a product explainer illustration, not a screenshot of a workflow editor. Browser/app/computer execution must be the dominant scene.
+
+Composition: wide landscape, approximately 16:9, high resolution, intended to remain legible when displayed around 1100 pixels wide. Quiet dark charcoal background. Four clearly readable stages across the image, connected by a few clean, short directional arrows. The middle work area should have the most visual weight and width. A subtle rounded enclosing surface visually unifies the team, its tools, and results as one employee operating system. Do not crowd the margins.
+
+Top: refined white serif wordmark text "OpenCompany" at modest size, followed by a large clean sans-serif heading "How AI employees work". Subtitle, exactly: "An operating system for AI employees". Do not invent a logo symbol.
+
+Stage 1, heading "1 Give a job": a compact owner message card, with the exact text "Research competitors and prepare a brief." Use a small person/profile line icon. The job clearly flows into the employee team.
+
+Stage 2, heading "2 Employee team": an elegant lead card with a purple circular avatar and label "Lead", coordinating two smaller specialist cards labeled "Research" and "Writing". Make their relationship clear with delicate connectors. A small checkmark near the lead communicates review. The team is software agents, not humans sitting at desks.
+
+Stage 3, heading "3 Do the work": the largest area, showing a realistic but simplified OpenCompany workspace panel with the micro-label "LIVE WORK". Within it, three distinct visible capabilities, each with a large readable title:
+- "Browser": a miniature browser window with tabs, a search field and clean research-result/document shapes; no dense text.
+- "Apps & integrations": a calm row/grid of generic envelope, calendar, document and repository line icons, showing connected services; no fabricated brand marks.
+- "Computer & tools": a small file-and-code workspace with document/file shapes, a few abstract code lines and a terminal icon. Convey reading/writing files, running code and using computer tools.
+These are active ways to complete the job, not merely lists of connectors. Use subtle activity marks/checks, not a glowing science-fiction network. Keep all inset UI large enough to understand.
+
+Stage 4, heading "4 Deliver results": a Workspace output card with the exact title "Competitor brief", a document/report preview using abstract chart and text blocks, and a green checkmark with the pill label "Ready". Nearby include a small understated card labeled "Review & feedback" to show the owner can inspect and coach. The completed artifact must be visually distinct from the initial job.
+
+Bottom: an elegant feedback path connects the delivered work and owner feedback back toward the employee team. Along that path place two compact reusable-resource cards with the exact labels "Memory" and "Skills", with database/notebook and playbook line icons. A short caption, exactly: "Carry knowledge and instructions into the next job". Make the feedback direction unambiguous. This represents retained knowledge and editable skills, not training the underlying AI model.
+
+Palette grounded in onboarding: application background #0d0f13; panel #15171c; elevated surfaces #1b1e25; borders #2b2f37; primary text #e8eaed; muted text #9aa1ac. Restrained agent purple #bd93f9, browser/interface cyan #8be9fd, tools/results green #50fa7b, optional very small warm-orange feedback accent #ffb86c. Do not use every accent equally. No heavy neon glow.
+Typography: Geist-like clean modern sans-serif for all labels and copy; small monospaced uppercase only for LIVE WORK; white serif only for the OpenCompany wordmark. Excellent hierarchy and readable type, no tiny decorative microcopy. Render all quoted wording accurately and do not add other text.
+Avoid: robots, humanoid employee cartoons, gears, conveyor belts, workflow-node palettes, cables everywhere, n8n-style node graphs, dark cyberpunk, stock-photo people, fake product claims, automatic-model-retraining imagery, watermarks, extra brand logos, dense paragraphs, busy dashboard filler. This must communicate an OS that equips AI employees to complete real work across browsers, apps and the computer.
+
+Input-channel edit prompt:
+
+Use case: precise-object-edit.
+Edit the supplied OpenCompany explainer image. Make one targeted change: add the proper job input channels inside the existing leftmost "1 Give a job" panel.
+
+Keep the overall image dimensions, framing, layout, background, title, subtitle, serif OpenCompany wordmark, all existing stages 2-4, work windows, report, arrows, memory/skills feedback path, colors, typography and visual style unchanged. Do not redesign or rearrange the overall image. Only change the inside of stage 1, whose panel occupies approximately x=20..252 and y=258..714 in the 1672x941 source.
+
+Keep the heading "Give a job" and the example "Research competitors and prepare a brief." The example can be made a little more compact within its current panel if necessary. Use the existing empty lower portion of this panel for a clean, readable input-channel group.
+
+Add a small uppercase micro-label "INPUT CHANNELS", then six tasteful compact chips arranged in two columns and three rows, in this exact order:
+Row 1: "Chat" and "Email"
+Row 2: "WhatsApp" and "Telegram"
+Row 3: "Discord" and "Schedules"
+Under them, add one full-width compact chip labeled exactly "Webhooks (Dev mode)".
+
+Each chip has a clear little line icon appropriate to the channel: chat bubble, envelope, phone/chat, paper plane, community chat, clock/calendar, and webhook connection. Generic channel icons are fine; do not add inaccurate brand logos. Use the same calm charcoal elevated cards, hairline gray borders, clean light sans-serif labels, subtle purple/cyan accents and restrained shadows as the existing image. All seven channel labels must be legible and spelled exactly as supplied. Keep the chips comfortably within the stage-1 panel, with consistent spacing and enough padding; no overlap with the example message, heading, outgoing arrow or border. Do not add descriptions or extra tiny text.
+
+Semantics: external messages or scheduled work can give an AI employee a job; webhooks are available in configured Dev-mode workflows. Do not show internal teammate task completion as an external input channel. Do not modify any labels, content, text or composition outside the stage-1 panel.
+
+Final app-tile edit prompt:
+
+Use case: precise-object-edit.
+Edit the supplied OpenCompany explainer image with ONE targeted change: make the apps recognizable in the "Apps & integrations" area.
+
+Change only the six app tiles inside the existing "Apps & integrations" inset panel, approximately x=815..1052, y=329..656 in the 1672x941 image. Keep the panel heading and its border. Replace the six generic unlabeled icons with six clearly recognizable real application tiles in the existing three-column, two-row grid:
+Top row: "Gmail", "Calendar", "Drive".
+Bottom row: "Sheets", "GitHub", "Stripe".
+
+Each tile must show the recognizable app logo above its exact readable name. Use Gmail's multicolor M-envelope; Google Calendar's blue calendar with 31; Google Drive's green/yellow/blue triangular mark; Google Sheets' green document/spreadsheet mark; GitHub's white Octocat silhouette; and Stripe's purple wordmark or stylized S. These are supported integrations, not made-up applications. Use proper recognizable app logos, keeping them small enough to fit cleanly. The Calendar/Drive/Sheets names refer to Google's apps. Do not change their labels to longer names that would crowd the tiles. Use clean light sans-serif labels and generous icon-to-label spacing. Preserve subtle green connection dots and the calm dark elevated-card styling, hairline borders, and restrained shadows. Make the tiles and names crisp and legible.
+
+Critical invariants: do not modify the overall dimensions, composition, wordmark, heading, subtitle, "Give a job" panel, any of its seven input channels, employee-team cards, Browser panel, Computer & tools panel, report, feedback card, memory/skills path, or any text outside these six app tiles. Keep every input-channel label, including "Webhooks (Dev mode)", intact. Do not add new sections or enlarge the integration panel. No redesign, no new artwork outside this one grid.