alec foster subscribe

lab

The lab.

Twelve interactive components living on this site. They are the site: the closest thing I have to a portfolio of how I think about agentic UI, terminal aesthetics, and builder-facing tooling. Everything here is open source.

$ git clone github.com/alectivism/personal-website →

featured_trace_viewer

Press play to watch a mock audit_campaign agent task execute step-by-step. Each step lights up cyan while running, locks chartreuse when done.

audit_campaign running
taskaudit_campaign
orchestratorClaude Opus 4.7
workers0 / 5
elapsed0.0s
  1. [0] mcp.firecrawl.scrape · 0.0s
    $ mcp.firecrawl.scrape("ads.competitor.com/q4-launch")
    → fetching 14 pages...
    → extracted 42 ad creatives, 8 landing pages
    ✓ 1,247 tokens captured · pii-filter: pass
  2. [1] mcp.exa.search · 2.4s
    $ mcp.exa.search("brand voice Q3 competitor")
    → 23 results, top 7 selected
    → semantic similarity threshold: 0.78
    ✓ corpus assembled · 8,124 tokens
  3. [2] analyze_claims · 6.8s
    $ analyze_claims(corpus, n=247, model="claude-opus-4-7")
    → dispatched to 4 sub-agents
    → claude_a analyzing claims 1-62 ...
    → claude_b analyzing claims 63-124 ...
    → claude_c analyzing claims 125-186 ...
    → claude_d analyzing claims 187-247 ...
    ✓ 247 claims classified · 8 flagged for review
  4. [3] flag_risks · 9.1s
    $ flag_risks(threshold=0.7)
    → scoring 8 flagged claims against risk taxonomy
    ⚠ 2 high-risk: unsubstantiated comparison claims
    ⚠ 1 high-risk: implied health benefit
    ✓ 3 critical, 5 watch · queued for governance review
  5. [4] draft_report · 12.3s
    $ draft_report({format: "md", sections: 5})
    → assembling exec summary, findings, risk register, recommendations
    → audit log written: gov_log_2026_05_11_audit_142.jsonl
    ✓ report.md · 1,842 words · ready for review
// rai_monitor pii_filter ✓ pass alignment ✓ pass bias_monitor ✓ pass audit_log ✓ written

agent swarm

live

Live simulation of a multi-agent system: 77 simulated workers (Claude, OpenAI, Gemini, open-source) orchestrated against real task types. Click to spawn. Drag to dispatch. Hover to inspect.

homepage open →

ask alec

demo

Scripted Claude-Code-style chat with quick-reply chips and streaming answers. A demo of what a terminal-native UI feels like, not a real LLM.

homepage open →

command palette

live

Press ⌘K from anywhere to open. Fuzzy search across all pages, quick actions (toggle theme, copy email, switch palette), keyboard nav.

⌘K press ⌘K

agent-readiness quiz

live

Seven-question diagnostic that maps your org onto three tiers and recommends an engagement. No backend, no email required. Score lives in localStorage.

/assessment open →

interactive 404

live

Triggered on any missing URL. Real shell: type `cd <route>`, `ls`, `help`, `whoami`, or bare paths to navigate. The unhappy path I actually want people to see.

/404 open →

pricing tier toggle

live

Three-mode toggle (advisory / build / speaking) revealing 7 engagement options with $-scale ranges and pre-filled mailto links.

/work-with-me open →

animated counters

live

IntersectionObserver-triggered count-up for hero credibility numbers. Respects prefers-reduced-motion (skips to final value).

homepage open →

career timeline

live

Seven-entry vertical timeline with gradient rail, per-event dots, and 'now' highlighted in chartreuse.

/about open →