From sign-in to first report in 5 minutes.

A quick walk through your first session — sign in, pick an industry, run a meeting, watch the post-session pipeline produce a signed report.

This guide assumes nothing. No prior setup, no API keys, no command line. If you have a browser, a microphone (or a transcript you can paste), and a Google account, you have everything you need. If you get stuck, write to hello@felarity.com and someone who built the product will answer.

What you will end up with

By the end of this walkthrough you will have:

• A signed-in account on a personal workspace.
• One completed session — either a short live recording or a pasted transcript.
• One finished intelligence report with contradictions, speaker attribution, acoustic markers, topology classification, and an 8-node Ed25519 attestation chain.
• A verified attestation — the chain checked against our public key, from outside the app.

Total elapsed time: about five minutes, most of which is the meeting itself.

Step 1 — Sign in

Go to app.felarity.com and click Sign in with Google. Free and Starter accounts use Google OAuth. Professional and Enterprise workspaces can wire up SAML SSO instead — your admin will have already done that, and you will see your IdP's login screen rather than Google's.

On first sign-in we create a personal workspace named after you. You can rename it, invite teammates, or create additional workspaces later from the sidebar. Free accounts have one workspace and one seat.

Step 2 — Pick your industry

Felarity ships with thirty industry councils — board governance, M&A diligence, litigation, clinical care, regulatory affairs, sales, vendor management, and so on. Each council is a fixed roster of four AI analysts whose role descriptions, evidence weights, and disagreement thresholds are tuned for that domain. The council you pick determines who is in the room when contradictions get scored.

If you are not sure which to pick, use General. You can change the council mid-session; doing so re-scores the open contradictions against the new roster.

Step 3 — Start a meeting

Two ways in.

Live recording. Click Start meeting. Your browser asks for microphone permission; grant it. We capture five-second WebM chunks, prepend the WebM init header to each, and stream them to Whisper large-v3 for transcription. On Pro and Enterprise, pyannote speaker diarization runs alongside the STT pass, so segments arrive already tagged with speaker labels.

Paste transcript. Click Paste transcript and drop in any text you already have. Helpful when you are testing the product against a real prior meeting, or when audio capture is not allowed. Contradiction detection works the same way; only the acoustic markers stage of the post-session pipeline will be skipped.

Step 4 — Watch contradictions surface live

As soon as the council has enough context, contradictions start arriving in the right column as cards. Each card shows the two statements that disagree, a confidence score, the analysts who flagged it, and a one-line synthesis.

One thing worth pausing on: contradictions are detected before speakers are attributed. The live pass scores the language, not the person. Attribution happens in the post-session pipeline, after full-session diarization has reconciled overlapping turns. This ordering is deliberate — it keeps the detection step honest. We are not deciding what to flag based on who said it.

Step 5 — Stop the session

When the meeting ends, click Stop & save. This triggers the post-session pipeline — an SSE stream of eleven stages that runs server-side and reports progress live. Typical wall-clock time is one to three minutes for a thirty-to-sixty-minute session.

The eleven stages, in order: concatenate audio chunks, run full-session diarization, extract acoustic markers, classify confrontation responses, extract per-speaker voice samples, attribute contradictions to speakers, re-score every contradiction with DeBERTa-v3 NLI on CPU, build the contradiction graph and classify its topology, run a deep council pass with full speaker context, build the 8-node Merkle attestation chain, and generate the final report.

Step 6 — Read the report

The report opens with the executive summary at the top — the synthesis the council settled on after seeing the full transcript with speakers attributed. Below that, contradictions are grouped by speaker pair and ranked by confidence. Each entry shows both statements, the NLI score, the acoustic markers around each statement (speech rate, pause length, stress), and the council's commentary.

Further down you will find the topology section — Felarity classifies the contradiction graph into one of a handful of patterns (isolated, hub-and-spoke, mutual disagreement, cascade, and so on). The pattern tells you something the contradictions alone do not: whether you are looking at one person under pressure, two people in a structural fight, or a group converging on a story that does not hold together.

At the very bottom is the attestation block — the 8-node Merkle chain, its root hash, the Ed25519 signature, and the public key fingerprint. You can download the report as JSON, Markdown, or PDF.

Step 7 — Verify the chain from outside

The whole point of the attestation chain is that it can be checked without trusting the Felarity app. Download the report JSON, then post it to the public verifier:

curl -X POST https://api.felarity.com/v1/verify \
  -H "Content-Type: application/json" \
  --data-binary @report.json

The response tells you whether each of the eight nodes hashes back to the root the signature covers, and whether the signature is valid against our published public key. You can pull the public key directly:

curl https://api.felarity.com/v1/verify/public-key
# or, equivalently:
curl https://app.felarity.com/.well-known/felarity-signing-key.pem

Both endpoints return the same Ed25519 public key in PEM form. If you want to do the verification entirely offline, save the PEM and use any standard Ed25519 library — Python's cryptography, Node's crypto, Go's crypto/ed25519, or openssl pkeyutl -verify.

If something goes wrong

Common first-session snags:

• No microphone prompt. Your browser may have remembered a denial. Open the site permissions and clear it, then reload.
• Live transcription falls behind. Whisper runs on a dedicated GPU and almost never lags, but a flaky uplink will. Drop to a wired connection if you can; the chunks will catch up.
• Diarization shows everyone as Speaker 1. Diarization is Pro and Enterprise only on live capture; on Free and Starter you can still paste in a transcript with speaker labels and the pipeline will use them.
• The post-session pipeline stalls. Each stage has its own timeout and will surface a clear error if it fails — there is no silent retry that hides a problem. If you see a stage stuck for more than two minutes, screenshot it and send it to hello@felarity.com.