Product

Inside Signals.

Run one research brief across every frontier model, check each finding against the open web, and keep the workspace live as the world changes.

Walkthrough

From a research question to a workspace you keep open.

Phase 1

Get set up

A workspace, a clear question, and a first run across the models.

01 · Workspace

Open the app, the work's already moving.

The moment you sign in, Signals hands you a workspace: the home for one research program. A few sessions are already going, so you're not staring at an empty screen wondering where to begin.

  • The workspace holds everything: your runs, findings, structure, and brand.
  • Each session is its own line of research. Pick one up and keep going.
  • Everything you do from here builds up in the same place, run after run.
Envisioning/Horizon Q
+New
Horizon Q3 sessions
Session α — Market landscape42 signals · updated 2h ago
Session β — Competitor watch28 signals · updated 1d ago
Session γ — Regulatory shifts16 signals · updated 3d ago
02 · Brief

Pin down the real question first.

A short chat works out the scope, the audience, the time horizon, and what to leave out. That brief drives every step after it, and you can edit it later and run again.

  • The chat draws out your objectives, your scope, and what to skip.
  • Every later step traces back to the brief, so the answer stays on the question.
  • Anyone with access can edit it, and the old versions stick around.
Envisioning/Horizon Q/Session α/Briefing
+New
Session α — draft brief
Autonomous agents in B2B procurement
Briefing assistant
What change in your industry over the next few years would matter most to your strategy?
Reply…
03 · Run

Put every frontier model on it at once.

Each run sends your brief to the big models from OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, Meta, Alibaba, and Moonshot. When their answers overlap, we fold them into one finding and note which models agreed.

  • Several models at once beats one model working alone.
  • Overlapping answers get folded into a single finding.
  • You can see where the models landed on the same answer on their own.
Envisioning/Horizon Q/Session α/Generate
+New
Generating signalsacross 9 models
0/9 ready
OpenAI
GPT-4.1
Anthropic
Claude Opus
Google
Gemini Pro
xAI
Grok
DeepSeek
DeepSeek R1
Mistral
Mistral L
Meta
Llama
Alibaba
Qwen
Moonshot
Kimi
Phase 2

Trust the findings

Check each finding, read it your way, and pull the best together.

04 · Verify

See the evidence behind every finding.

A finding is more than a paragraph. It comes with its sources, a verdict on each one, and the reasoning behind how confident it is. We check the claims against the open web, and anything that contradicts them stays on the record.

  • Each source is marked as supporting, contradicting, unrelated, or unreachable.
  • Confidence comes from those verdicts, not from the model's own say-so.
  • When the sources disagree, the finding says so.
Envisioning/Horizon Q/Session α/Synthetic agents
+New
Artificial Intelligence
Synthetic agents
Autonomous software actors that negotiate, plan, and act on behalf of organizations across the open web.
VerifiedGrowthEarly majority
05 · Explore

A radar you actually work in.

Findings land on a radar: quadrants for your categories, rings for whatever metric you pick, like readiness, impact, or urgency. It's how you move around the work, not a chart you save for the last slide.

  • Your categories, metrics, and tags set the layout, defined once for the workspace.
  • Click to read, filter to narrow it down, or copy the URL to share the exact view.
  • The same findings also read as a table, a card grid, or one detailed view.
Envisioning/Horizon Q/Session α/Radar
Grid
Radar
+New
06 · Curate

Cherry-pick findings across every run.

Findings don't have to stay in the run they came from. A set pulls the ones you want from anywhere in the workspace, shaped for a particular story or audience.

  • Pick findings from across many runs and gather them into one set.
  • A set points at findings rather than copying them, so nothing drifts out of sync.
  • Brief an exec, prep a workshop, or seed the next run.
Envisioning/Collections/Favorites
+New
AI
Synthetic agents
Energy
Autonomous grid
Bio
Cellular agriculture
Society
Civic AI tribunals
Materials
Bio-glass facades
Mobility
Robotaxi cohorts
🔍Add to collection…
Favorites13
Near-term pilots4
Moonshots7
Ops brief — Q13
+ New collection
Phase 3

Keep it alive

The workspace compounds over time, and you can reshape it anytime.

07 · Compound

It gets sharper every time you re-run.

Runs stack up inside the workspace. Re-run the brief next quarter and the radar shifts: findings get added, verified, revised, or retired. The activity feed stamps every change with a time.

  • Categories, metrics, tags, and workflows stay on the workspace across every run.
  • The activity feed shows what was added, verified, revised, and retired.
  • You set the cadence: weekly, monthly, quarterly, or whenever you want.
Envisioning/Horizon Q/Session α/Live
+New
Live·updated continuously
Activity
Sources:9 models
08 · Revise

Change your mind, reshape the radar.

After a run, tell the analyst chat what to change: merge categories, raise the bar for verification, narrow the time horizon. The radar rebuilds as you go, and you can still trace where everything came from.

  • The assistant can edit findings, sources, reports, and sets for you.
  • While a run is going, edits are held so nothing collides.
  • Every theme and report can be rebuilt from the findings underneath.
Envisioning/Horizon Q/Session α/Revise
+New
Revise radar
Deepen the AI wedge — more agentic-economy signals.
Revise the radar…
And then it runs itself

It keeps watch after you close the tab.

Pick a cadence and Signals re-runs the brief on its own, nightly, weekly, or monthly. It keeps an eye on your sources, catches what changed, and updates the radar without you asking.

  • Re-runs on a schedule: nightly, weekly, or monthly.
  • Between runs it flags new, moved, and dead links.
  • Only what changed gets touched. The rest of the workspace stays put.
Envisioning/Horizon Q/Session α/Automation
+New
Automation3 schedules active
Nightly digestevery day · 02:00
scheduled
Weekly re-runMondays · 09:00
scheduled
Monthly deep scan1st · 06:00
scheduled
Next runwatching sources
Same engine, per question
Seven scan types, each shaped for a different kind of research question.
Pick the scan that fits →
Workspaces

Make it look like your team built it.

You get your own brand, categories, metrics, and radar shape. The same engine runs underneath, but nothing about it looks off-the-shelf.

  • Pick your accent, dark surface, and radar geometry.
  • Set the workspace up once and every run inherits it.
  • Publish a run publicly, keep it private, or share read-only.
Envisioning/Horizon Q/Session α/Radar
Grid
Radar
+New
Envisioning
Built for enterprise

Get it past your security team.

Bring your own AI gateway key, your own SSO provider, and your own access policy. Signals stays out of your auditor’s way.

Bring your own AI key

Run on Vercel AI Gateway with your own keys and rate limits. You pay the models directly, so the AI bill from us is €0, and the audit trail stays in your workspace.

SSO and roles

Sign in through your own provider, scope access to your organization, and hand out owner, editor, and viewer roles. Runs and findings all follow the same rules.

Public, private, or embargoed

Publish a run as a public page, keep it private to your org, or share a read-only set with one partner. Every finding carries its own visibility.

Benchmark · May 28, 2026

32 frontier models tested on 12 industries.

We grade every model on the same axes we use inside the product: verifiability, specificity, currency, and coverage. Open methodology, open data, refreshed quarterly.

Top model
Claude Opus-4.7
Anthropic · 86.0 composite
Models tested
32
Industries
12

See it on a question you actually have.

One run on our credits. It becomes the first run of your workspace if you keep going.