Skip to content

armgit / AiSOC

Public Imported from GitHub
0

Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.

4.6 KB
Raw Blame

90-second AiSOC demo screencast — shot list

Phase 4.3 / T6.4 — the canonical script for the public product walkthrough.

If you're recording the screencast, follow this file frame-by-frame. If you're editing the marketing site to embed the cut, the file path the player loads is apps/web/public/demo/demo.mp4 (master) and apps/web/public/demo/demo-poster.png (poster). The recorder pipeline in .github/workflows/screencast.yml writes both to that directory.

Total target runtime: 90s ±2s. Anything beyond 92s gets re-cut.

Pre-roll (0:00 — 0:02)

  • AiSOC wordmark fades up on a dark background.
  • No voiceover. No music sting. Two seconds, dead silent — sets up the "we're going to show you, not pitch you" frame.

Shot 1 · Alert hits (0:02 — 0:14)

  • Visual: browser at localhost:3000/dashboard/alerts. A new alert card animates in from the top (the seed dataset includes a Crowdstrike EDR detection for wmic.exe spawn from winword.exe).
  • Voice (12 words): "An EDR detection comes in. AiSOC opens an investigation automatically."
  • Cut on: the alert card finishing its slide-in animation.

Shot 2 · Auto-investigation timeline (0:14 — 0:30)

  • Visual: click into the alert. Investigation view opens. The timeline ticks through: enrichment → triage agent → correlation → host graph. The narration agent panel shows three sentences of natural language summary.
  • Voice (15 words): "The triage agent enriches the event, queries the host graph, and writes a plain-language summary."
  • Cut on: the narration agent finishing the third sentence.

Shot 3 · Tool calls + audit trail (0:30 — 0:44)

  • Visual: scroll the right rail. Every tool call the agent made is listed with arguments + result. Hover over crowdstrike.isolate_host shows the audit-trail expandable card with the OPA decision JSON.
  • Voice (13 words): "Every tool call is logged with arguments, result, and the OPA policy decision."
  • Cut on: the OPA decision JSON visible for two seconds.

Shot 4 · Human-in-the-loop approval (0:44 — 0:58)

  • Visual: the agent proposes isolate_host. Approval modal appears. Show the inline diff (host → isolated), the playbook reference, and the "Approve" button. Click Approve.
  • Voice (16 words): "High-impact actions need a human. The modal shows what will change before you say yes."
  • Cut on: the modal dismissing.

Shot 5 · Action executor (0:58 — 1:10)

  • Visual: the timeline shows the executor card running. Status goes Pending → Running → Success. The host card updates to "Isolated" with the Crowdstrike host ID and a link out to the EDR console.
  • Voice (14 words): "AiSOC fires the action against Crowdstrike, records the response, and updates the host."
  • Cut on: the host card showing "Isolated".

Shot 6 · Compliance + report (1:10 — 1:20)

  • Visual: click the "Generate report" button. The compliance dashboard tile updates ("1 contained · 0 escalated"). The PDF report opens in a side panel.
  • Voice (12 words): "Every step is captured for the SOC 2 evidence trail. One click. Done."
  • Cut on: the PDF cover page in frame.

Outro (1:20 — 1:30)

  • Visual: fade to the AiSOC wordmark on dark with "MIT licensed · github.com/beenuar/AiSOC" under it. No deployment hostname — a URL in the outro dates the cut and points viewers at one particular instance rather than at the project.
  • Voice (9 words): "Run it yourself: one command, no account required."

Recorder pipeline

  1. The CI workflow .github/workflows/screencast.yml runs on a manual trigger only (recording is a deliberate human act, not a per-push side effect).
  2. The workflow spins up a Playwright runner against a pinned dataset (apps/web/e2e/fixtures/demo-seed.json) so the alert that animates in is always the same wmic.exe detection. Without a pinned seed the screencast would drift on every record.
  3. The runner navigates through Shots 1–6 with hand-tuned timing constants in apps/web/e2e/demo/screencast.spec.ts and writes demo.webm + a poster frame.
  4. ffmpeg transcodes the webm to mp4 (H.264 / AAC) at 1080p for browser compatibility, then uploads both as workflow artifacts. A separate PR commits them to apps/web/public/demo/ so the marketing site can serve them.

Why this lives in docs/demo/ (not the marketing repo)

The script is engineering-owned because every shot maps to a specific code path (which means a refactor that moves the host-graph panel silently breaks Shot 2). Pinning the shot list next to the code keeps that link visible.