v0.7.2

Features

• Restore Code Activity analytics for modern Codex (>= 0.149.1) rollout format, including file edits, lines changed, language breakdown, reads, and searches. #386
• Derive a missing Codex first_user_message from the first transcript chunk so sessions remain visible in the session list, Trends, and cost-by-model. #385

Pricing corrections and coverage

• Correct Sonnet 5 and fast-mode pricing, remove retired date-based Sonnet pricing, and add verified pricing for previously unpriced Anthropic, OpenAI, Gemini, Grok, Mistral, and DeepSeek model families. #383
• Update GPT-5.6 Terra and Luna pricing to reflect the July 30 OpenAI price reduction. #382
• Add missing Claude Opus 5/Mythos 5 and OpenAI GPT-5.3 through GPT-5.6 pricing entries. #381

Operations

• Add opt-in DB_CONN_MAX_IDLE_TIME and configure production for Neon autosuspend. #380
• Reduce Fly.io app and worker memory allocations from 1024 MB to 512 MB. #379

Refactoring

• Remove backend OpenTelemetry tracing and its unused dependency surface. #384

Docs

• Add an AGENTS.md symlink to the repository development guidance.

DB Migrations

None.

API Changes

• POST /api/v1/sync: for Codex transcript uploads, the server now derives an omitted metadata.first_user_message from the first chunk's event_msg user message. A client-supplied value continues to take precedence.
• POST /api/v1/admin/cards/invalidate: documentation now includes the already-supported session_card_tokens_v2 card type for pricing-card invalidation.

Breaking Changes

• The backend no longer emits OpenTelemetry spans. Deployments that depend on OTLP/Honeycomb tracing must migrate to structured logs or remove their tracing configuration. No public HTTP routes, request fields, environment-variable names, or CLI commands were removed or renamed.