REM From FunnelScript

REM researches how a company acquires customers from a public website, App Store listing, or Google Play listing. Each run saves a Markdown report and a JSON evidence bundle. The CLI, local web app, and MCP server use the same collectors and run history.

Install

Download the archive for your platform from GitHub Releases. Compare its SHA-256 hash with SHA256SUMS, extract it, and put rem or rem.exe on your path. The macOS archives contain unsigned binaries.

With Go installed, you can build and install directly:

go install github.com/funnelscript/rem-from-funnelscript/cmd/rem@v0.1.0
rem demo

Run it

From the source directory with Go 1.26 or later:

make build   # or: go build -trimpath -o rem ./cmd/rem
./rem doctor
./rem linear.app --only website,seo --no-ai --out ./rem-reports/linear
./rem validate ./rem-reports/linear/evidence.json ./rem-reports/linear/report.md
./rem web

Build & test automation

make build    # Build optimized binary with -trimpath and -ldflags="-s -w"
make test     # Run full test suite with race detector (-race)
make lint     # Run go vet ./...
make smoke    # Run end-to-end smoke test on linear.app (or $REM_LIVE_TARGET)
make install  # Install binary to $(GOPATH)/bin/rem
make cross    # Cross-compile for linux/amd64 and darwin/arm64 into dist/

./rem demo opens a recorded Cal AI investigation with no requests to source providers. Use --no-open --port 0 to print a dynamically allocated loopback URL without launching a browser. Choose Evidence only in the browser to collect and view reports without a model.

To compare two snapshots:

./rem linear.app --only website,seo --no-ai --json
./rem linear.app --only website,seo --no-ai --json
./rem diff linear.app

rem diff also accepts two run IDs, two evidence files, or two report directories. An item absent from a later checked sample is described as no longer seen in that sample. It is not proof that an advertiser stopped running it.

Sources and coverage

Area	Evidence collected
Website and funnel	Homepage messaging, CTAs, pricing and trial signals, linked social accounts and apps, affiliate and team pages, logo candidates
Tracking	Public page and bounded Google Tag Manager tag signatures, attribution and marketing stack
SEO and AI search	Robots rules, llms.txt, nested and gzip sitemaps, URL families, locales, and update dates
History	Wayback snapshot availability and sampled changes in homepage messaging
App stores and launches	App Store metadata and ratings, Google Play public install bands, Apple EU Ad Repository, Hacker News
Founders	Public website hints and matched Wikidata claims. Optional paid provider enrichment for LinkedIn and X
Paid and social	With Scrape Creators, Meta, Google and LinkedIn ad samples, TikTok, Instagram, YouTube, X and LinkedIn, Reddit and Product Hunt

Collectors label unavailable data as unknown and include manual-check links where possible. Tracking tags signal a possible channel. Ad samples and public install bands do not establish conversions, ROAS, spend, or causal impact.

--only accepts website,seo,history,apps,paid,social,founders,community. The website stage always runs when a public website is available. For store input the store lookup also runs first. Use --handle tiktok=example to override a discovered handle. Default exports go to rem-reports/<run-id>. Pass --out to choose a directory. If no observations are collected, the CLI exits nonzero and retains the run's source statuses and unknowns.

Optional providers

• SCRAPECREATORS_API_KEY (alias SCRAPE_CREATORS_API_KEY) enables paid-source collection. --max-credits caps dispatched requests per run (default 50, range 0 to 500). The JSON credit count reports request reservations, which includes failed dispatched requests. Provider billing can differ. Cached successful responses may be up to 24 hours old and do not use a request reservation. Set REM_NO_CACHE=1 before a comparison that requires fresh paid-source observations.
• ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN enables the Anthropic writer. --engine claude-code and --engine codex use the installed and authenticated CLI. --model can override the selected writer's model. --no-ai and CLI --json use the deterministic evidence writer.
• Strategy writers receive the evidence needed to draft the report. A failed request or validation falls back to a saved evidence report and records a writer warning. Reports and provider data are local unless you choose an external writer. Configure credentials through your shell environment.

History defaults to the operating system's user config directory under reverse-engineer-marketing-go/runs. Set REM_HOME or pass --home to use another location. The local web app listens on IPv4 loopback. It exposes history, run progress and cancellation, report and evidence downloads, and content review. Jobs are local to the running process. Completed runs persist across restarts.

Use with an MCP client

./rem setup --dry-run
codex mcp add reverse-engineer-marketing-go -- /absolute/path/to/rem mcp

Use the binary's stable absolute path. The same stdio command works with other clients:

{"mcpServers":{"rem-go":{"command":"/absolute/path/to/rem","args":["mcp"]}}}

MCP offers investigation, evidence retrieval, content filtering, playbooks, report validation, report writing, and run comparison. Use get_evidence with include_unknowns: true to retrieve unresolved checks. The named prompts cover company investigation, comparison, monitoring, planning, and unknowns.

Contribute

See CONTRIBUTING.md for development and release checks, CHANGELOG.md for changes, and SECURITY.md for private security reporting.

Verify

go mod verify
go vet ./...
go test -race -count=1 -cover ./...
go build -trimpath -o rem ./cmd/rem
bash smoke.sh ./rem

For live verification use REM_LIVE_TARGET=linear.app bash smoke.sh ./rem. Direct outbound DNS and HTTPS are required by the public fetcher. Cross-compilation examples and platform checks are recorded in the verification ledger.

This is a new Go implementation with an original browser interface. Compatibility is measured by the workflow and evidence contracts in the parity matrix. API field shape and collection limits can differ from the TypeScript release. Paid API and authenticated strategy adapters have fixtures and require real credentials for live confirmation.

The upstream project, embedded analyst playbooks, and recorded Cal AI demo are by Daniel Hangan (DansUGC), under MIT. See LICENSE and the pinned reference.