Skip to main content

Release 1.2.0: Chat History, Message Editing, and Audit Accuracy

Kevin Kantesaria
Kevin Kantesaria
Erik Lutz
Erik Lutz
Sri Aradhyula
Sri Aradhyula
Naren Ambati
Naren Ambati

Highlights​

CAIPE 1.2.0 brings a unified chat history, message editing with conversation rewind, and a context-remaining indicator to web chat, plus native rich-text support for Jira. It also widens team-based management permissions and fixes a batch of authorization audit gaps that were undercounting or hiding real activity.

What's New​

Chat​

  • Edit and rewind messages — Edit a message you already sent and rewind the conversation to that point; the original history stays intact if you change your mind (#2781).
  • Context-remaining indicator — Web chat now shows how much context budget is left before the conversation compacts, on by default and configurable in Settings (#2779).
  • Unified conversation history — Chats started from the browser, API, autonomous tasks, schedules, and webhooks now appear together in one history list, with faster incremental loading and a resizable sidebar (#2778).

Jira​

  • Native rich text — The Jira integration can now write real Jira formatting — headings, links, lists, code blocks, and tables — into comments and descriptions instead of flattening it to plain text (#2766).

Authorization & Audit​

  • Team-based management — Team members, not just team admins, can now manage knowledge bases, RAG collections, MCP tools, and Webex spaces owned by their team, matching how agent and Slack channel ownership already work (#2773).
  • Faster catalog lists — Agent, MCP server, and workflow config/run lists now determine what you can see with a single lookup instead of one check per item in the whole catalog, so these pages stay fast as the catalog grows (#2793).
  • Smaller audit trail — Routine authorization allows are aggregated into periodic counts instead of one row per check, cutting audit storage and query costs. Denials are always kept per-event, and a new setting can restore full detail for a bounded investigation window.

Bug Fixes​

  • Admin Insights: human activity, active-user counts, and conversation totals are now counted consistently across browser, Slack, Webex, and direct API chat, so people who only used a bot connector or the API weren't missing from the numbers (#2780).
  • Audit evidence: bulk authorization checks — like rendering the agents list — are now recorded as one event with accurate allow/deny counts instead of one row per resource. Exported audit evidence previously undercounted these decisions (#2777).
  • Bedrock model calls: fixed calls to Claude Sonnet 5, Opus 5, Fable 5, and Mythos 5 through Bedrock, which were failing outright because of a sampling parameter these models reject (#2776).
  • Platform-health audit: fixed a TLS scheme mismatch that silently broke the platform-health notification audit behind an ingress (#2752).
  • Release notes: the in-app release notes load from the correct repository again (#2761).
  • Docker Compose first install: the RAG stack's Milvus dependency no longer fails to pull on a fresh install (#2769).
  • Helm image pulls: the OpenFGA authorization bridge chart can now pull its image using a Kubernetes image pull secret (#2398).

Security​

  • RAG's own authorization checks — search, per-tool, per-datasource, and org-admin approvals — are now captured in the audit trail. Previously these decisions were made but never recorded.

Breaking Changes​

No breaking changes. Drop-in upgrade from 1.1.1.

Known Issues​

None known at this time.

Upgrade Guide: 1.1.1 → 1.2.0​

Overview​

Drop-in upgrade — no values.yaml edits required. Two new optional Helm values ship with defaults that preserve existing behavior.

Helm Values Changes​

New Optional Fields​

KeyDefaultDescription
openfga-authz-bridge.imagePullSecrets[]Image pull secrets for the OpenFGA authorization bridge Deployment, for private registries
openfga-authz-bridge.audit.fullFidelityAllowsfalseStore one row per authorization allow instead of periodic aggregate counts. Enable only for a bounded investigation or compliance window
openfga-authz-bridge.audit.allowRollupFlushSeconds60How often aggregated allow counts flush to the audit trail

Breaking Changes​

None.

Deprecated / Removed Keys​

None.

Upgrade Runbook​

1. Update the chart​

helm upgrade ai-platform-engineering \
  oci://ghcr.io/caipe-io/charts/ai-platform-engineering \
  --version 1.2.0 \
  -f your-values.yaml

2. Apply values.yaml changes​

None required. If you need full-fidelity audit rows for authorization allows during a compliance review, set:

openfga-authz-bridge:
  audit:
    fullFidelityAllows: true

3. Verify​

kubectl get pods -n <namespace>