changelog

Changelog

What we shipped, and when.

v0.7latest

May 2026

Semantic codebase model + cross-org intelligence

Caret now builds a semantic model of your codebase and uses cross-organization intelligence to surface patterns, drift risks, and improvement suggestions before your agent even starts.

Semantic codebase model — AST-aware file graph with dependency tracking
Cross-org intelligence network — anonymized pattern sharing across organizations
Rule engine v2 — custom drift rules with YAML-based configuration
Network client for real-time pattern exchange
Codebase health scoring with risk heatmaps
Improved CLI with `caret model` and `caret network` commands
Supervisor v2 with context-aware drift thresholds
Marketing site redesign — warm palette, scroll-driven terminal demo

v0.6

May 2026

Supervisor management + health tracking

Manage supervisors as first-class resources. Health tracking, daemon process management, and real-time supervisor status.

Supervisor management dashboard
Daemon process for background supervision
Health tracking with trend analysis
Notification system (email, Slack, desktop)
Activity feed with real-time updates
Weekly digest reports

v0.5

May 2026

Agent Supervisor — real-time drift detection

The big one. Caret now supervises your AI agents in real-time, detects when they drift from your intent, and intervenes before damage is done. Phase 1 ships watchdog mode with five drift detectors.

Real-time supervisor via Claude Code hooks (PreToolUse, PostToolUse, UserPromptSubmit)
Intent extraction — locks what you asked for, evaluates every action against it
Destructive action detection — catches rm -rf, force push, DROP TABLE
Retry loop detection — flags agents stuck in error-fix-error cycles
Scope drift detection — alerts when the agent edits unrelated files
Cost runaway detection — warns when token spend spirals
Error spiral detection — catches cascading failures early
Desktop notifications on drift via macOS native alerts
Session state persistence at ~/.caret/supervisor/
Health scores per session
Supervisor install/uninstall/status commands
Run comparison page
Cost analytics command in CLI
Global search in command palette (Cmd+K)

v0.4

May 2026

Interactive CLI shell

Run Caret in supervision mode. The CLI now supports an interactive shell that continuously supervises runs without restarting.

Interactive shell mode with `caret shell`
Real-time run streaming to dashboard
Configurable supervision filters
Shell history and auto-complete

v0.3

April 2026

Pattern matching + cross-team learning

Caret now identifies recurring failure patterns across your organization and surfaces fixes that teammates have already found.

Automatic failure pattern detection
Cross-project pattern matching
Suggested fixes from team history
Pattern dashboard with trend charts

v0.2

March 2026

Run summaries + onboarding flow

AI-generated summaries for every run, plus a guided setup experience for new users.

AI-generated run summaries on ingest
Guided onboarding flow for new workspaces
Improved run detail view with timeline
Cost breakdown per run

v0.1

February 2026

Initial release: capture, replay, dashboard

The first version of Caret. Capture any CLI agent run and replay it in the browser.

CLI capture with `caret run`
Web dashboard with run timeline
Full run replay: stdout, stderr, git diff
Project management and API key generation
Basic cost tracking via token counts

Semantic codebase model + cross-org intelligence

Caret now builds a semantic model of your codebase and uses cross-organization intelligence to surface patterns, drift risks, and improvement suggestions before your agent even starts.

Semantic codebase model — AST-aware file graph with dependency tracking

Cross-org intelligence network — anonymized pattern sharing across organizations

Rule engine v2 — custom drift rules with YAML-based configuration

Network client for real-time pattern exchange

Codebase health scoring with risk heatmaps

Improved CLI with `caret model` and `caret network` commands

Supervisor v2 with context-aware drift thresholds

Marketing site redesign — warm palette, scroll-driven terminal demo

Supervisor management + health tracking

Manage supervisors as first-class resources. Health tracking, daemon process management, and real-time supervisor status.

Supervisor management dashboard

Daemon process for background supervision

Health tracking with trend analysis

Notification system (email, Slack, desktop)

Activity feed with real-time updates

Weekly digest reports

Agent Supervisor — real-time drift detection

The big one. Caret now supervises your AI agents in real-time, detects when they drift from your intent, and intervenes before damage is done. Phase 1 ships watchdog mode with five drift detectors.

Real-time supervisor via Claude Code hooks (PreToolUse, PostToolUse, UserPromptSubmit)

Intent extraction — locks what you asked for, evaluates every action against it

Destructive action detection — catches rm -rf, force push, DROP TABLE

Retry loop detection — flags agents stuck in error-fix-error cycles

Scope drift detection — alerts when the agent edits unrelated files

Cost runaway detection — warns when token spend spirals

Error spiral detection — catches cascading failures early

Desktop notifications on drift via macOS native alerts

Session state persistence at ~/.caret/supervisor/

Health scores per session

Supervisor install/uninstall/status commands

Run comparison page

Cost analytics command in CLI

Global search in command palette (Cmd+K)