Tools: what fills each role without a person
Every role a freelancer fills can also be filled by a tool. This is the list, with what each one costs per asset and whether its output can be checked by a machine.
3,652 products806 gigs priced across 13 roles800 chart rows, 13 categories245 listings with multiples150 categories ranked11 pages rendered from the index20 posts published by machine
| tool | kind | fills | step | interface | deterministic | unit cost | status | notes |
|---|---|---|---|---|---|---|---|---|
| Ayrshare | api | outreach | publish | http | yes | $149 | archive | one API for X/TikTok/YouTube/Instagram/Reddit/Bluesky — the multi-rail outbox in one call, but $149/mo is 15% of the budget before revenue |
| HOPE | estate-repo | clipper | publish | cli | yes | to measure | archive | harvest once, publish everywhere — the multi-rail publish step, to revive |
| Postiz | library | outreach | publish | http | yes | $0 | archive | schedules to X, YouTube, TikTok, Instagram, Threads, Bluesky…; still needs our own X app keys; heavy to host |
| crx-corpus | estate-repo | researcher-verifier | scan | cli | yes | $0 | archive | Chrome Web Store scraper/index — the data product for the extensions niche (copy, do not rebuild) |
| omni-extension | estate-repo | outreach, community-mod | publish | extension | no | to measure | archive | universal browser automation: Chrome extension + CF Worker orchestrator + local daemon — the base for our own extensions |
| agentkey | api | researcher-verifier | scan | api | yes | to measure | candidate | live data for agents: web, X, Reddit, YouTube, finance, e-commerce; price unmeasured |
| aisa | api | run | api | no | to measure | candidate | one key for models + data + skills with per-call billing and budget caps — the budget cap is the interesting part for an unattended tick; price unmeasured | |
| anthropics/skills | oss | distribute | repo | yes | $0 | candidate | 176,141 stars — the venue for the credential (decision 11): the number gate as a skill, listed in awesome-claude-skills (74,970) | |
| browser-use | oss | scan | python | no | $0 | candidate | 114,523 stars — agents that use the browser; vs our cdp.mjs / chrome-js.mjs (deterministic, no model). Python: say why before adopting | |
| claude-mem | oss | memory | plugin | no | $0 | candidate | 93,810 stars — persistent context across sessions; overlaps our memory dir + records-as-text; trial against, do not add on top | |
| crawl4ai | oss | researcher-verifier | scan | python | yes | $0 | candidate | 83,256 stars · 8,596 forks · Apache-2.0 · created 2024-05-09, pushed 2026-09-09. URL in, clean LLM-ready Markdown out; no key, no signup, no per-page pricing. Rail 2 for sources with no clean payload — the free answer to firecrawl (180,004). Python: adopt only where our .mjs scanners cannot read the page. |
| firecrawl | oss | scan | api | yes | $0 | candidate | 180,004 stars — search/scrape/context API; our scanners are cheaper and deterministic; candidate only for sources without a clean payload | |
| langfuse | oss | observe | self-hosted | yes | $0 | candidate | 34,552 stars — traces, evals, cost per run. Today: loop-log.md + hermes cron runs + post.cost_usd by hand | |
| orca | oss | orchestrate | app | no | $0 | candidate | 67,847 stars, Mar 2026 — an ADE for a fleet of parallel agents; the owner asked for agents aware of each other (Claude Code + Antigravity + Hermes) | |
| promptfoo | oss | gate | cli | yes | $0 | candidate | 25,069 stars — test prompts and agents; would turn the number gate into a regression suite over past pieces | |
| rtk | oss | desk | cli-proxy | yes | $0 | candidate | 80,200 stars, Jan 2026 — proxy cutting LLM tokens 60–90% on dev commands; for the interactive sessions, not the tick | |
| tikhub | api | researcher-verifier | scan | api | yes | to measure | candidate | public social data API: TikTok, Douyin, Xiaohongshu, Instagram, X, YouTube, Bilibili, Weibo, Kuaishou, Reddit, LinkedIn — recent posts per account, batch public comments, creator lists. Rail 2 (paid API) for platforms we have no scanner for; price unmeasured |
| uumit | marketplace | distribute | web | no | to measure | candidate | A2A marketplace: register agents, APIs, data and skills as capability cards that other agents find, call and pay for — where the Skool parser or the index could be LISTED (archetype 33) | |
| Google Flow / Veo | subscription | faceless-video | render | extension | no | to measure | paused | estate repo: Chrome extension + server for programmatic Flow control — catalog scan, media harvest, tab routing (plate-harvester agent exists) |
| Remotion | library | faceless-video, video-editor | render | cli | yes | $0 | paused | estate repo: Layer-3 Remotion compositor driven by a schema — the deterministic render for evergreen assets (script + plates → mp4); gate = transcript diff |
| UGC-generator | estate-repo | face-presenter | render | cli | no | to measure | paused | AI UGC pipeline for TikTok affiliate — the AI-avatar credibility tier |
| backlot | estate-repo | faceless-video | author | cli | yes | to measure | paused | video production as a document the system compiles — 35 films one format |
| clip-rankings | estate-repo | clipper | score | cli | yes | $0 | paused | clip ranking studio + competitor vault (irreplaceable local data) |
| replicator | estate-repo | author | cli | no | to measure | paused | AI workflow studio; prior attempt at the same OS — mine for parts, do not resume | |
| youtube-api | estate-repo | researcher-verifier | count | cli | yes | $0 | paused | the publish+count step's official read; replaces page scraping for our own channels |
| Typefully | subscription | outreach | publish | http | yes | $12.5 | pending-owner | posting without our own X app: outbox → Typefully API → scheduled threads; they carry the X API cost |
| Whop Trends Pro + MCP | subscription | researcher-verifier | scan | mcp | yes | $29 | pending-owner | shortcut for daily coverage of every category; our scanner is the free substitute |
| X API pay-per-use | api | outreach | publish | http | yes | $0.015 | pending-owner | the sanctioned posting path; follow/unfollow/like writes are Enterprise-only since 2026-04-20 (not on pay-per-use) — the unfollow chore cannot go through the API |
| Apify | api | researcher-verifier | scan | http | yes | to measure | to-wire | Substack/Skool/Gumroad/YouTube scrapers as actors — the rails chrome-js does not reach at scale |
| ElevenLabs | subscription | voice | render | http | yes | to measure | to-wire | replaces the $68-avg voice gig; deterministic given voice_id + text; gate = transcribe back and diff (memory: transcribe-generated-audio-to-verify-it) |
| Google AI Studio / Gemini | subscription | researcher-verifier, designer | author | http | no | to measure | to-wire | text + image generation; thumbnails via image model; non-deterministic → always behind a rubric gate |
| Browser MCP | mcp | outreach, community-mod | publish | mcp | no | $0 | working | clicking/typing in Chrome; not deterministic (accessibility snapshot), so used only where a URL or count confirms the action |
| Claude Code | subscription | researcher-verifier, outreach | author | cli | no | to measure | working | the author and the desk; subagents for collect/organize/rank; cost is the machine line in cash_ledger |
| cdp.mjs | cli | researcher-verifier, outreach | scan | cli | yes | $0 | working | a SEPARATE Chrome per brand (own user-data-dir under ~/.cache/x-profiles), launched with open -g (never foreground), driven over DevTools Protocol; navigator.webdriver=false; measured 2026-09-13: reads x.com logged-out; a logged-in brand session needs one login by the owner in that window |
| chrome-js.mjs | cli | researcher-verifier | scan | browser-js | yes | $0 | working | JS in the owner's logged-in Chrome via Apple Events; ~0.35 s per read; the read side of every scanner |
| hermes-agent | oss | run | cli+gateway | no | $0 | working | 245,172 stars — runs our tick (operator-tick, operator-scan) and holds the gateway; #2 agent repo on GitHub, which is itself a credential | |
| scan-fiverr.mjs | cli | researcher-verifier, outreach | scan | cli | yes | $0 | working | 806 suppliers across 13 roles in one pass |
| scan-whop.mjs | cli | researcher-verifier | scan | cli | yes | $0 | working | 150 categories · 3,504 products · enrichment from whop.com landing pages |
| simulate.mjs | cli | score | cli | yes | $0 | working | 120 months per concept from rate rows | |
| twitterapi.io | api | researcher-verifier | scan | http | yes | $0.00018 | working | the read instrument for the 10,000-profile sample and follow graphs: ~$2 for 10k profiles, no session, no browser, no foreground · key saved 2026-09-13 as TWITTERAPI_IO_KEY in fbi-vault/.env; calibrated on @whotfiszackk: API 5,791/9/1,155 = browser read |
| vidmatic | estate-repo | faceless-video, video-editor | render | cli | yes | to measure | working | video replication monorepo (active); replicate-beat skill; corpus + reference side-by-side |
"Deterministic" means the same input gives the same output and a gate can check it; anything else sits behind a rubric.