Join the waitlist
Platform · AI controls and spend

Your AI, your rules.Down to the file path.

Choose which model does each job, what Tore fixes and when, how far it goes on its own and what it may spend. Or leave it on Auto, with the reason for every choice shown.

Join the waitlistTour every setting

Tore opens soon. Join the waitlist and we'll email you the day you can start.

app.tore.ai/beacon-data/settings/ai
Example data

Example: the AI settings for the Beacon Data workspace, with groups for model routing, what it fixes, when it works, autonomy, budgets, risk and hand-off, and notifications. An admin changes the model for grouping errors from Grok Fast to Gemini Flash, raises the severity threshold to high and above and adds Chrome extension errors to the ignore list, moves the beacon/web repository from assist to auto-draft while billing stays locked to observe, and turns on slowing down near the monthly cap instead of stopping. Each change shows a saved confirmation.

Granular when you want it.Sensible when you don't.

Leave it on Auto and Tore picks a recommended set-up, with the reason for each choice. Or open any group and set it down to the repository, the path and the hour.

The right model for each job

Fast models answer quick questions, deep reasoning takes the hard bugs, coding models write the fixes. If a provider goes down, the fallback takes over.

One model for everything

Only the work you want

Set the products, repositories, paths and severity it picks up. Ignore the noise, and fence off the files it should never touch.

An AI chasing every warning

Autonomy that grows with trust

Start one repository on observe and move another to auto-draft. The risky parts of your code always go to a person.

All or nothing automation

No surprise AI bill

Every AI action is metered per product, with caps, alerts and a kill switch. Hit zero and support keeps working.

Finding out from the invoice

Every setting, group by group.Change as much or as little as you like.

Model routing

Pick the model for every task, or let Tore pick.

Sorting a message and rewriting a payment path are different jobs. Set the model family and reasoning level per task, a fallback for when a provider is down, and a second model from a different vendor to check every fix.

  • Model and level per task: answers, triage, root cause, the fix, the review
  • A fallback model per task, so support keeps running through an outage
  • One switch: let Tore route everything, with the reason for each choice
  • A second model reviews each fix before your engineer sees it
Early access

Bring your own AI keys during early access. Early teams can plug in their own Claude, OpenAI, Grok or Gemini keys while Tore still routes every task.

app.tore.ai/beacon-data/settings/routing
Example data
Let Tore route everythingRecommended set-up, reason shown for each task
Answer a how-tofast · fallback Claude HaikuGemini Flash low
Find the root causedeep reasoning · fallback Gemini ProClaude Opus high
Write the fix and its testcoding · fallback Claude OpusCodex Astra high
Second review of every fixmust be a different vendorGemini Pro med
Why Codex Astra here: fixes need a coding model that runs your tests. Claude Opus takes over if it's unavailable.

What it fixes

Decide exactly what it picks up.

Point Tore at the products and repositories you want covered, set how serious a problem must be before it acts, and filter the noise. The paths it should never touch stay off limits.

  • Rules per product, repository, source and environment
  • Severity and impact thresholds before anything acts
  • Pick the error types it works on, ignore the rest
  • An ignore list, an always-escalate list and never-touch paths
app.tore.ai/beacon-data/settings/scope
Example data
Repositories and pathsproduction only
beacon/web · src/**beacon/api · services/**beacon/mobile
SeverityHigh and above, or 10+ customers
LowMediumHighCritical
Error types
TypeError5xx responsesTimeoutsDeprecation warnings
Ignore
Script error.Bot traffic
Always escalate
checkout/*Enterprise
Never touch
infra/terraform/**src/legal/***.env*

When it works

Choose what starts the work, and when it runs.

A customer report, a spike in errors, a failed test run or a feature request that crosses a vote count can each start Tore on its own. Set working hours, let it run overnight, and cap how much it does at once.

  • Triggers: customer reports, error spikes, failed tests, request votes
  • Working hours per team, plus overnight runs ready by morning
  • A limit on how many investigations run at the same time
  • Hourly and daily action limits, plus a stop switch
app.tore.ai/beacon-data/settings/schedule
Example data
Triggersstart an investigation when…
A customer reports a bugchat widget, email, Slack
Errors spike50+ events in 10 minutes
A test run failson main
A feature request reaches 25 votesbuild it as a change to review
Working hours9:00 to 18:00 ET
MonTueWedThuFriSatSun
Overnight runsfindings ready by 8:00
Investigations at once-3+

Autonomy and risk

Set how far it goes, repository by repository.

Each repository gets its own autonomy level, and each area its own approval rung. By default fixes wait for your engineer's approval. Six risk classes always go to a person, and you can add your own.

  • Four autonomy levels: observe, assist, auto-draft, verify on staging
  • Four approval rungs per area: automatic, confidence-gated, approval required, human review
  • An area's setting can only be tighter than the default, never looser
  • Add your own risk classes for the code only you know is sensitive
app.tore.ai/beacon-data/settings/autonomy
Example data
beacon/web
ObserveAssistAuto-draftStaging
beacon/api
ObserveAssistAuto-draftStaging
Approval rungs in beacon/apitighter only
src/ui/**Automatic
services/search/**Confidence-gated
services/export/**Approval required
services/payments/**Human review · billing
Risk classes6 locked · 1 yours
AuthBillingMigrations Personal dataTenant isolationSecrets Report exports

Budgets and limits

Know what it costs before finance asks.

Every AI action is metered and attributed to a product. Cap each investigation and each month, get alerts on the way up, and choose what happens near the limit. One switch stops all AI work, and support keeps running.

  • Caps per investigation, per day and per month
  • Spend per model and per product, in a usage report with plan limits
  • Near the cap: slow down on lighter models, or pause
  • A kill switch. At zero, the inbox, help center and boards keep working
app.tore.ai/beacon-data/settings/budgets
Example data
October AI spend$238 of $400
Alert at 80% to #beacon-engPer investigation $4.00 cap
By model
Claude$112
Codex$71
Gemini$38
Grok$17
By productBeacon Web $151 · Beacon Mobile $87Usage report
Near the capSlow downPause AI
Stop all AI workSupport keeps running without itKill switch

Hand-off and step-in

A person can step in at any moment.

Set the confidence below which Tore hands a conversation to a person, and the topics where it always offers one. Anyone can take over a thread, request changes on a fix in plain words, or correct an answer so Tore gets it right next time.

  • A confidence threshold for handing a conversation to a person
  • Hand-off rules: when, whether and for whom to offer a human
  • Take over any thread, or request changes on a fix and Tore re-runs the tests
  • Correct an answer or a diagnosis, and it learns from the correction
app.tore.ai/beacon-data/settings/hand-off
Example data
Hand to a person below70% confidence
Always offer a person:Billing questionsEnterprise accounts
Maya · Northwind

Can you move our seats to the annual plan mid-cycle?

Tore: confidence 58% · handing to a personTake over
Hana took over · Tore's notes kept in the thread
Fix TCK-482 · request changesrevised
Alex

Keep the 1,000 row cap for free workspaces, and add a test for it.

- const limit = PAGE_MAX;
+ const limit = ws.plan === "free" ? 1000 : PAGE_MAX;
Tore

Revised. Tests re-run: 49 of 49 passing, 1 new.

Hana corrected an answer: export limits apply per workspace, not per user. Tore will use this next time.

Who gets told

Choose who hears about what.

Send each kind of event where it belongs: fixes waiting for approval to the right approvers, risk-class changes to a channel, budget alerts to the owner. The customers who reported a bug hear back when the fix is live.

  • Per event: Slack channel, email or the in-app feed
  • Approvers per repository and per risk class
  • Reporters told automatically when their fix is live
  • Reply and act from Slack, without a second tab open
app.tore.ai/beacon-data/settings/notifications
Example data
EventSlackEmailIn app
Fix ready for approval
Risk class touched
Root cause found
Budget at 80%
Handed to a person
ApproversSlack: #beacon-eng
Alex beacon/web Dan beacon/api Sarah billing class
Tell reporters when their fix is liveIn the conversation they started
6risk classes locked to a person, in every repository
4autonomy levels, set per repository
2models on every fix: one writes it, a different vendor checks it
1switch to stop all AI work, with support still running

Keep exploring.

Capture & fixFixes and approvalsThe fix arrives tested and reviewed. Your engineer says yes.See how → ProductTore AIOne AI team for support and engineering, on the right model for each job.See how → Capture & fixAutopilotLet Tore pick up errors and reports on its own, inside your rules.See how →

Start on Auto.Tune it when you're ready.

Tore picks a sensible set-up on day one. Every setting on this page is there when you want it.

You're on the list. We'll email you the day you can start.

Get early access when Tore opens. No spam.