How it works
Paste companies, a CSV, or an Apollo/ZoomInfo/CRM export. Or drop a GitHub repo and DevSignal reads the real companies behind the stars. No CRM connection, no setup.
Every company checked for real, graded on real need from public evidence, with the one dated timing signal found or an honest nothing. Junk dropped with a cited reason.
You get the verified buyers, the one move for each, and the message drafted. Mark what converts, and the next read is sharper.
Bring a scraped list, a CSV, or an Apollo/ZoomInfo/CRM export. DevSignal reads it, drops the junk (universities, OSS repos, no-fit companies) each with a cited reason, then grades who actually needs you and dates the timing signal. The receipt leads with what it killed, and the outreach you would have wasted.
Toggle to Contacts and the same flow re-verifies an exported contact list: buyer titles, company fit, and whether the email domain accepts mail.
Open any account and DevSignal shows the read: why they fit with cited reasons, the score and exactly how it was computed, a dated activity timeline, and the play that worked for a similar company. Then the one move, with the message drafted and the warmest person to send it to.
A thin account gets an honest hold, not a fabricated draft. One click prints the whole thing as a one-pager.
Tell it what you want in plain words, or pick a saved job: vet a list, research an account, find buyers, re-verify your pipeline, position against a rival. It plans the steps, you confirm, then it runs the same verified engine.
It never free-roams. Every step is grounded, so a hallucinated name gets killed with a reason instead of landing in your pipeline.
Point DevSignal at a prospect's public GitHub org and it quantifies the unmerged, unreviewed work: open PRs by age, time to first review, time to merge, and the stale dependabot and renovate PRs, foregrounded. Every number is dated and links to the GitHub query it came from.
Two tiers, always labelled. Measured, with real numbers from a real org, or Inferred, when public data cannot measure it. Never an estimate dressed up as a fact.
Model engineer-hours saved, cost per merged ticket before versus after, monthly net saving, and the acceptance rate you have to clear to break even. Every intermediate number shows the arithmetic that produced it, including the AI inference cost.
Pure math, no fabricated numbers, and the breakeven point is drawn on a curve, not asserted. Share a scenario as a read-only link.
The whole cockpit
Re-check a CRM or Apollo export: which titles are not buyers, which companies do not fit, and which email domains accept mail. Honest about the ceiling, it confirms the domain, not a specific mailbox.
Your whole pipeline, honestly: fit mix, intent tiers, the funnel, recurring themes, contact reachability. It does not guess industry, revenue, or headcount.
Diagnose your metrics (LTV:CAC, CAC payback, NRR, runway) with the classic traps flagged, plus a browsable library of GTM frameworks.
Your won customers and the ones named on your own site, plus one real, nameable contact per account. Never a fabricated stakeholder.
Track a rival: who adopts their OSS, the comparison threads, the problem-space discussions, and how to position against them.
One compound row per account, ranked by conviction: the forks, the hiring, the competitor moves, and the public triggers, stacked with a plain why-now.
Ask who is warmest, what changed since last time, or how to approach an account. Answers cite the evidence, and threads persist.
A daily Slack ping when a tracked account heats up or a new signal lands. The one integration that is live today.
Every won and lost outcome sharpens the next read, pooled and anonymised across customers. No public source has this. No stateless agent session can build it.
A bigger scraped database goes stale the day the data is wrong. A record of what actually closed only gets sharper.
It starts as an honest heuristic from fit and conviction, and turns into real rates once your outcomes fill the pool. It always shows the sample size behind the number.
For agents
The same verification engine installs as an MCP server, the standard way to give an AI agent new tools. Claude, Cursor, or any MCP client can call its six tools while it works, so it stops guessing and starts checking. And when an agent researches companies through your tools, that demand is yours to see, captured from your own key, never scraped. The engine and your keys stay on our side.
The six tools your agent gets
How it works
The whole install, one block:
{
"mcpServers": {
"devsignal": {
"command": "node",
"args": ["./devsignal/index.js"],
"env": {
"DEVSIGNAL_URL": "https://devsignal-rho.vercel.app",
"MCP_API_KEY": "your-key"
}
}
}
}The GTM companion that verifies and remembers. Fit-graded, cited, dated.