What's new

Changelog.

What shipped, what changed, what's next. Tracked publicly.

· Aug 2026beta

›Evidence freshness: an approved action is re-checked against current evidence before it executes — consent is no longer sufficient authority on its own
›Approvers see drift before deciding: pending cards pre-flight the check and name exactly what moved (status transitions, vanished records) instead of a generic warning
›Materiality filtering: only changes bearing on the action block it — unrelated project churn is reported without holding the action, so the gate stays worth reading
›Acknowledged overrides are sealed in the hash-chained OAR envelope, so replay distinguishes “nothing changed” from “a human saw the change and accepted it”
›Honest verdicts: a comparison basis older than the freshness threshold returns “cannot confirm” and blocks rather than giving a false all-clear; a failed check also blocks
›Evidence-gate effectiveness stats — catches and override rate together, so a gate that starts crying wolf shows up as a defect
›CRITICAL FIX: GitHub PRs and issues had two identity schemes (poller keyed on number, webhook on global id), so one object became two rows and a merge was invisible to agents and to any freshness check
›CRITICAL FIX: the GitHub poller requested only open PRs/issues and hardcoded status, so merged and closed transitions never landed without a webhook configured

· Apr 2026beta

›Start free, no credit card — full platform access on signup (750 messages, all connectors)
›Team plan: organizations, RBAC, invite flow, multi-seat workspaces
›Audit log export — CSV + JSON download for SIEM integration
›Dedicated pricing page with full tier comparison
›Comparison pages: Pantheon vs CrewAI, Pantheon vs Lindy
›Use-case pages: /for/engineering, /for/marketing, /for/operations
›Design partner application page with structured intake form
›Day-2 reactivation emails for dormant trial users
›Onboarding simplified from 5 steps to 2 steps (80% activation target)

· Apr 2026beta

›CRITICAL: AsyncLocalStorage for tenant isolation (fixes cross-tenant data leak)
›CRITICAL: OAuth redirect uses request origin (fixes staging→prod redirect)
›CRITICAL: 30s per-round timeout on tool execution (fixes agent hangs)
›CRITICAL: Memory load errors surfaced to agent (fixes silent fail-open)
›CRITICAL: ErrorBoundary at root + OS desktop level (fixes full whitescreens)
›Anthropic prompt caching — 40% cost reduction on LLM calls
›Sentry error tracking (backend + frontend)
›CI/CD pipeline with backend + frontend jobs
›API key scope enforcement (read/write/admin)
›Request-level 2-minute timeout middleware
›Per-tenant rate limiting (30 req/min on message endpoint)
›Health check verifies DB connectivity (returns 503 on failure)
›Webhook dead-letter queue with exponential backoff retry
›Structured logging with request IDs on every log line
›Standardized connector error handling with 15s timeouts
›Graceful WebSocket reconnection with exponential backoff
›Load testing baseline script (bun run test:load)
›117 automated tests (up from 80) — plan limits, RBAC, validation, tenant isolation

· Mar 2026beta

· Jan 2026alpha

· Dec 2025alpha

· Nov 2025alpha