TIPHYSby Intelligant AIBook a demo
Η΄The ship's log

What's new in Tiphys

Release notes, new capabilities, and notable fixes. Tiphys was called Clu before September 2026.

Tiphys

  • Clu is now Tiphys, by Intelligant AI. Same product, new name.
  • The site moved to tiphys.intelligant.ai. getclu.dev and clu.intelligant.ai redirect here and keep the path, so existing links still work.

Marketplace listing

  • Flat-rate AWS Marketplace pricing — $0.10/hr per pod-hour while Clu is running. Single Marketplace dimension; every capability (Core observability, Core Plus writes, Cloud, Developer Platform) is included at the flat rate.
  • 30-day free trial activates automatically on first subscribe — no charges until it ends.
  • Roughly $72/cluster/month for a single-pod install running 24/7. Scale to zero to pause billing.
  • Which capabilities you turn on is a Helm-values decision (modules.core.enabled, modules.corePlus.enabled, modules.cloud.enabled, modules.developerPlatform.enabled); the bill doesn't change based on which ones are on.
  • Replaces the earlier six-dimension tier-package model — that model was never customer-facing publicly, and AWS Marketplace's flat-rate container product type only exposes a single hourly dimension. Existing v0.2.x code paths work against the flat rate without further changes.

v0.2.5

  • Helm install via the Marketplace ECR registry is now the canonical and only install path. The previous AWS-console / aws eks create-addon delivery option is retired — install with helm install clu-ops-agent oci://709825985650.dkr.ecr.us-east-1.amazonaws.com/cloudology/clu-ops-agent ... against your EKS cluster.
  • Flat top-level Helm values for the four install parameters (service_account_role_arn, bedrock_region, marketplace_tier, prometheus_url) so a single --set per knob configures the install.
  • Customers running v0.2.0–v0.2.4 should upgrade — earlier versions in the v0.2 line had install gaps that v0.2.5 fully closes.

v0.2.0

  • AWS Marketplace listing live with hourly metering — aws-marketplace:MeterUsage reports usage once per hour. Best-effort: failed reports log a warning and retry on the next tick. Scale to zero pauses billing.
  • Capability gating — which capabilities run is resolved from Helm values at every hourly check; no pod restart needed when values change.
  • Pre-v0.2.0 customers running v0.1.x see no behavior change on upgrade.

v0.1.12

  • Module wire-keys renamed to align with the new tier names — modules.cluster.* → modules.core.* and modules.idp.* → modules.corePlus.* in Helm values
  • Display names follow the same rename: "Cluster Agent" → "Core", "IDP Platform" → "Core Plus"
  • Backwards-compatible: JWT licenses with old module names still validate via a legacy alias map. Operators with custom Helm values overrides need to rename to the new keys.

v0.1.11

  • Bedrock cost tightening — chat token spend roughly 70-85% lower with no quality regression
  • Default chat model flips to Haiku 4.5 (Sonnet remains opt-in for hardest-case multi-hop planning via agent.model.interactive: smart)
  • Tighter LLM-bound conversation history (last 10 turns; older turns stay in the UI store for display) plus an 8 KiB cap on individual tool results
  • Rolling summarization of dropped older turns so the agent keeps context across long sessions without paying full input-token cost on every turn

v0.1.10

  • Approve button now shows a tool-aware progress label during dispatch ("Helm installing…", "Applying manifest…", "Calling AWS…") instead of sitting silently disabled for 60+ seconds on long installs
  • Reject and Edit-plan buttons hide while a dispatch is in flight so the only on-screen interaction is the live progress indicator

v0.1.9

  • "Send feedback" link in the in-cluster sidebar — opens a pre-filled email to support@cloudology.cloud with version, current route, and active modules so context lands with the report. No telemetry, fully user-initiated.
  • Cloud Agent IAM expanded with iam:SimulatePrincipalPolicy so the pre-flight permission check doesn't fail closed on otherwise-valid writes the operator already approved

v0.1.8

  • Agent-driven installs of cluster baseline components (metrics-server, pod-identity-agent, EBS CSI driver, etc.) now flow end-to-end through the approval gate so the agent can close the loop after recommending what's missing
  • AWS API responses with timestamp fields no longer crash chat with a JSON-serialization error

v0.1.7

  • Cloud capability gains IAM permission to install missing cluster baseline components (metrics-server, pod-identity-agent, EBS CSI driver, and similar) through the approval gate — when the agent recommends installing what's missing, it can actually install it instead of just naming the gap

v0.1.6

  • Smarter onboarding for new platform users — when a user signals inexperience ("I'm new to platform", "what's best", "what would you recommend"), the agent picks recommended defaults (ingress-nginx, ECR, EKS Pod Identity, namespace-per-team) and explains each in one sentence instead of presenting menus the user can't yet evaluate
  • Cluster discovery now runs before architectural recommendations — the agent checks what's already running in the cluster before suggesting installs, so it can't recommend something that's already in place

v0.1.5

  • Helm-install reliability — fixes the path that broke helm install --repo URL for every chart on first try
  • Enterprise / private chart-registry support via a new helm.repos[] Helm values block — wire internal HTTPS or OCI chart repos with credentials, no out-of-band setup needed

v0.1.4

  • Critical fix — approved helm install operations now reach the cluster. A production wiring bug was silently routing them to a no-op path; audit logs claimed success while the release never landed.
  • Approval card now surfaces the real outcome of each dispatched write — Applied, Approved-but-didn't-land, or Rejected — instead of always reporting success
  • Eleven common public chart repos are pre-registered at startup so helm install <alias>/<chart> works on first call
  • Cluster name flows from Helm values into the agent so it can pre-fill --name on EKS API calls without asking

v0.1.3

  • Critical fix — Marketplace RegisterUsage was failing parameter validation on every install. v0.1.2 and earlier subscribers came up healthy but ran with no capabilities active. v0.1.3 corrects the call and adds a regression test so this can't silently revert.

v0.1.2

  • Pilot-readiness patch — Clu is in preview and available exclusively on AWS Marketplace
  • Bedrock prompt caching cuts warm-conversation input-token spend to ~12% of prior
  • Tool-result size cap removes the runaway-bill tail risk on oversized payloads
  • Approval flow no longer leaks implementation language into chat
  • In-product copy uses Cluster Agent / Cloud Agent / IDP Platform names

v0.1.0

  • First public release on AWS Marketplace — preview / pilot phase, EKS only
  • Unified, classified tool surface for AWS, Kubernetes, and Helm operations
  • Support for non-Anthropic models on Bedrock (Llama, Mistral, Cohere)
  • Destructive operations (delete, uninstall, drain, uncordon) gated behind the same approval flow as other writes
  • Topology view rework — workload chains, status-driven coloring, drill-in navigation