Now
Calibrating for first cohort
You are here
The platform is running. We are configuring it for the first wave — tuning trust scoring thresholds, security scanning rules, and agent team compositions based on real project types.
Now in private beta → Central oversight for
Central oversight for
autonomous agent teams
Every tool call. Every file write. Every bash command. Watch your Claude Code agents work, approve what matters, and let trust scoring handle the rest. Finally get some sleep.
DevelopmentTimeline
Mar 14
Onboarding + first-run experience
Zero to running in minutes
Setup wizard walks through project connection, hook installation, and first escalation. Demo mode lets you explore with simulated data before connecting a real project.
Mar 13
Teams + roles + permissions
Built for collaboration
Invite team members with owner, admin, member, or viewer roles. Granular permissions control who can approve escalations, manage projects, or view analytics.
Mar 12
Analytics dashboard
Operational intelligence
Four KPI cards, daily task completion charts, escalation breakdowns, and cost trends. Shareable links for stakeholders. CSV export for everything.
Mar 10
Shared rule library
Reusable approval policies
Create, share, and import rule packs — "Python testing basics", "Security lockdown", "Documentation safe zone". Rules apply across projects and teams.
Mar 9
Session replay + audit trail
See exactly what happened
Full timeline of every action, escalation, and decision in a session. Filter by type, task, or agent. Expand any event for complete detail. Nothing happens without a record.
Mar 8
Mobile approval + notifications
Approve from your phone
Critical escalations push to your phone as notifications. Lightweight mobile interface for quick approve/reject decisions. Install as a PWA — no app store required.
Mar 6
Overnight mode
Agents work while you sleep
Tasks classified as green, amber, or red by risk level. Enable overnight mode and safe tasks run unattended. Pre-flight checklist verifies everything before you step away.
Mar 5
Session digests + changelogs
Automated reporting
Morning and evening summaries of everything your agents did. Weekly changelogs group work into Added, Changed, Fixed, and Removed — generated automatically from session history.
Mar 4
Cost visibility + budgets
Know what you're spending
Token usage and cost tracked per session, task, and project. Set daily spend limits with automatic alerts at 80% and hard stops at 100%. Full cost export for accounting.
Mar 3
Three-tier memory
Agents that remember everything
Core memory holds permanent project knowledge. Working memory tracks the current session. Episodic memory compresses history into searchable weekly summaries. Nothing is lost.
Mar 2
Agent template library
21 specialized agent roles
Plug-and-play agent teams for coding, research, document, and presentation projects. Each role has scoped tool access, security rules, and domain-specific methodology.
Mar 1
Trust scoring engine
Agents that earn their autonomy
Every approval builds trust. After consistent good behavior on a tool-file pattern, the system auto-approves — no more interruptions for decisions you've already made.
Backend API
38actions
31auto-approved
1hard stop
6done
$0.43cost
Current task
task 4Implement JWT auth middleware
bash: pip install pyjwt --index-url http://...
hard stop · waiting on decision
4s
Write src/auth/jwt.py
auto-approved · write
12s
Read requirements.txt
auto-approved · read
14s
Git checkpoint — 6 files changed
auto-approved · checkpoint
2m
Task 4 started — JWT auth middleware
PM agent
14m
Needs decision1 hard stop
Package install — unknown registryhard stop
Your agent wants to install a package from an unknown registry.
pip install pyjwt --index-url http://pypi.internal.io/simple
What led here
Read requirements.txt
Write src/auth/jwt.py
bash: pip install pyjwt —index-url...
Up next
Write token refresh logicready
Add auth middleware to routesready
Run test suitewaiting
Overnight mode off
Everything you need to oversee autonomous agents
Multi-project
One Dashboard, Every Project
Register coding, research, document, or presentation projects. Each gets its own agent team, memory tiers, and security profile.
4 project types
Per-project agent teams
3-tier memory architecture
Active projects
4 project typesPer-project agent teams3-tier memory
Trust scoring
Agents That Learn Your Preferences
The more you approve, the less you need to. Trust scores build per tool, per file pattern, per project — so safe actions auto-approve.
Pattern-based auto-approval
5+ approvals → auto-trust
One-click trust reset
Trust scores
Pattern-based auto-approval5+ approvals → auto-trustOne-click reset
Real-time oversight
Watch Every Agent Decision
Every tool call from every agent flows through your dashboard. Approve, reject, or let trust scoring handle it automatically.
Live WebSocket event stream
Per-event security scanning
Context-aware risk levels
Live event stream
Live WebSocket streamPer-event scanningContext-aware risk
Security scanning
Catch Risks Before They Ship
Every tool call is scanned for secrets, destructive commands, protected file access, and scope creep — before it executes.
Secret detection
Protected file enforcement
Scope creep alerts
Security findings
Secret detectionProtected file enforcementScope creep alerts
How it works
1
Install the hook
Run the install script on any project. It adds a lightweight PreToolUse hook that intercepts every agent action.
python hooks/install_hook.py --path ./my-project --type coding --goal "Build the API"2
Dashboard reviews
Events flow to your dashboard in real-time. Security scanning, risk scoring, and agent analysis happen automatically.
3
Agents learn
Every approval builds trust. After enough consistent approvals, the same actions auto-approve — no human needed.
See it in action
Agent performance
94%auto-approve rate
312actions this week
4.2tasks per session
See how your agents are improving
Track approval rates, task throughput, and session efficiency across all your agent teams in real time.
- Per-project success rates at a glance
- Identify bottlenecks before they slow you down
- Watch trust scores climb as agents learn your patterns
Know exactly what you're spending
Every API call is tracked and attributed to the right project. No more guessing where your Claude budget goes.
- Per-project cost attribution
- Daily and weekly spend trends
- Set budget alerts per project or globally
Cost breakdown
Total today$6.91
Simple, transparent pricing
Starter
$0
Free forever
For solo developers getting started with agent oversight.
- 1 project
- 100 events/day
- Basic trust scoring
- Community support
Pro
$29
per month
For teams running multiple concurrent agent projects.
- Unlimited projects
- Unlimited events
- Advanced analytics
- Shared dashboards
- Priority support
Enterprise
Custom
contact for pricing
Self-hosted deployment with full control and SLA.
- Self-hosted option
- SSO / SAML
- Audit log export
- Custom agent profiles
- Dedicated support
Frequently asked questions
A lightweight Python script runs as a Claude Code PreToolUse hook. Every time an agent calls a tool (file write, bash command, etc.), the hook POSTs the event to your dashboard. The agent blocks until you approve, reject, or the trust score auto-approves.
Auto-approved actions add <50ms of overhead. Escalated actions block until you decide — but as trust scores build, fewer actions need human review. Most teams see 90%+ auto-approval within the first week.
Four types out of the box: coding (frontend, backend, testing agents), research (data collection, analysis, summarization), document (writing, fact-checking), and presentation (slides, visual layout). Each type comes with pre-configured agent teams and security profiles.
Yes. The dashboard is a Flask backend + Angular frontend backed by PostgreSQL. Run it anywhere — your own server, a VM, or a container. No external dependencies beyond the database.
Every approval or rejection is recorded per (project, tool, file pattern). After 5 approvals with 0 rejections, that combination auto-approves. After 10 approvals, it's promoted to always-approve. 3 rejections → always-reject. You can reset any trust score at any time.
Ready to take control of your agents?
Set up in under two minutes. Watch every tool call, approve what matters, and let trust scoring handle the rest.