Core service

​

Durable billing and evaluation improvements

• Added the billing_reports table to preserve overage charges and retry Stripe reporting independently of billing-period closure.
• Fixed trailing usage blocking renewal, with duplicate-reporting safeguards and follow-up checks for delayed usage.
• Updated judge options to Gemini 3.5 Flash-Lite, Gemini 3.8 Flash, GPT-6 Luna, and Claude Haiku 4.5, and upgraded LiteLLM for GPT-6 support.
• Made evaluation-run names optional for selected traces and sessions, and moved the name field below the model selector in both creation sidebars.

​

Redis and billing reliability

• Hardened the Redis connection pool against stale Memorystore connections with health checks, TCP keepalive, socket timeouts, and bounded retries.
• Made usage counter increments and rollbacks replay-safe, preventing Redis retries from applying the same change twice.
• Added ownership-safe locks for Stripe webhooks and overage reporting so retries preserve idempotency without releasing another worker’s lock.

​

Billing, trace API, and dashboard improvements

• Updated invoice webhook handling for the pinned Stripe API version while retaining compatibility with the legacy invoice shape. Renewal periods and overage reporting now advance correctly.
• Changed PATCH /traces/{trace_id} to omit the potentially unbounded spans payload. GET /traces/{trace_id} continues to return the complete trace with spans.
• Added database connection pre-ping and recycling to avoid handing stale pooled connections to requests.
• Added an expandable span waterfall, fixed long JSON value truncation, and refined the dashboard’s dark theme and interaction states.
• Improved Agent Skill installation and onboarding guidance on the project home page.

• Refined task-completion scoring for traces that contain only Harness setup activity. These traces are now described as not having attempted the task yet, while subsequent task actions are judged normally.

​

Harness-aware evaluation

• Updated trace judges to recognize Harness meta-tools and treat guidance, state checks, notices, and trace inspection as neutral framework activity.
• Excluded Harness meta-tools from plan adherence, plan quality, step efficiency, argument correctness, tool correctness, and confidence penalties.
• Improved task-completion scoring for partial traces and Harness-only activity.

​

Worker and billing reliability

• Added worker-pool heartbeats to Cloud Run health probes and included prefetched tasks in queue-depth checks.
• Added dispatch locks, task time limits, and timeout handling for background jobs.
• Tightened Stripe client configuration and now fail fast when the Stripe API key is missing.

​

Agent-assisted onboarding

• Added project onboarding through the PandaProbe Agent Skill, with install commands and a ready-to-copy coding-agent prompt.
• Reorganized manual setup guides as secondary resources and refreshed the onboarding email flow.

​

CLI authentication

• Added OAuth 2.0 Authorization Code with PKCE endpoints, service logic, dashboard authorization page, and API integration for the PandaProbe CLI.
• Hardened token exchange by rechecking organization membership and retaining authorization codes when PKCE verification fails.

• Improved Celery’s Redis reliability with connection health checks, keepalive, timeout retries, and disabled unused task-result tracking.

​

In-product setup guides

• Replaced the standalone quickstart with a project sidebar containing guided setup for traces, agents, and evaluations.
• Added a dedicated API-key creation dialog and reusable instruction components.
• Improved trace and session tables by preserving the opened row and scrolling it into view.

• Expanded documentation for Deep Agents, LangChain, Amazon Bedrock, Mistral, and the Core and Python SDK changelogs.
• Updated Gemini model references and refreshed LiteLLM, FastAPI, asyncpg, pytest, Ruff, and release tooling dependencies.

• Fixed a translation-layer bug that could crash pages on load. Translations are now blocked by default to prevent the regression.
• Cleared a Radix Dialog warning that was surfacing in the browser console.
• Internal: added repo conventions (CLAUDE.md) and a formatting pass.

​

Org-scoped analytics

• Refactored the analytics service and the PostHog event taxonomy to be org-scoped instead of routing-scoped. Signup, API, eval-run, and trace events now all carry a consistent org identifier, making PostHog dashboards and funnels significantly cleaner.
• Added new events for org creation.
• Internal: tests updated for the new analytics shape; minor edge-case fix.

​

Project creation & onboarding UX

• Project creation moved to a dialog-based flow (matches the new API key creation UX in v0.1.9).
• Org creation now auto-provisions an onboarding project, so new orgs land in a usable state immediately.
• Removed the org-level quickstart page (now redundant with the onboarding flow).
• Updated the invitation email template.
• Onboarding checklist refinements.
• Internal: services migrated to use the svc layer instead of routing through repos directly, for consistency with the rest of the architecture.

​

API keys & dashboard polish

• New dialog-based API key creation flow with inline key display for the rotation feature.
• Added a shared Dialog component for reuse across the app.
• Dashboard formatting and theme polish; assorted bug fixes.
• Updated onboarding sample prompt.

• Fixed a Celery broker liveness probe issue that was causing intermittent worker restarts.
• Added an explainer video to the onboarding experience.

​

Onboarding & privacy

• Added an org-level quickstart page alongside the project-level quickstart.
• Onboarding status tracking; simplified onboarding checklist.
• Email and ID masking in PostHog events for privacy.
• Navigation polish: “back to home” button moved to the top, plus new CTA buttons and doc links.
• Surfaced LLM client key exports in the quickstart.
• Frontend formatting & test fixes; dead-link cleanup.

​

Documentation revamp

• Comprehensive documentation overhaul: quickstart, installation, welcome page, integrations overview, OpenAI / Anthropic / Google wrappers, LangGraph / CrewAI / Google ADK / Claude Agent SDK / OpenAI Agents SDK integration pages, eval intro / concepts / setup, session and user pages.
• New logo and updated header.
• Stripe promotion code support enabled at checkout.
• Fixed the user_authenticated PostHog event.

​

Public launch of the PandaProbe core service