Changelog

Release history for the AgentKavach SDK and platform.

v0.2.0 March 2026 #

Duration tracking, expanded budget types, and rate limiting improvements.

Duration Tracking

• New duration_ms field on ingest events — tracks LLM call latency in milliseconds
• Dashboard overview and agent detail pages now display cumulative duration
• Duration budgets are fully evaluable — alerts fire when duration thresholds are crossed
• Custom date range picker supports duration aggregation

Budget Types

• 7 budget types now supported: cost, tokens_total, tokens_input, tokens_output, calls, duration, and cost/monthly
• All 6 non-period budget types are independently evaluable for alerts
• Alert dedup key includes budget_type — same threshold fires once per type

Rate Limiting

• Simplified burst limits: Free=1K, Pro=5K, Max=10K events/second
• Removed dead RPM (requests per minute) code path
• Per-org rate limit overrides via admin API
• IP rate limiting simplified to flat 10K ceiling

v0.1.0 March 2026 #

Initial release of AgentKavach — hard budget limits for LLM APIs.

Python SDK

• Budget enforcement with daily, monthly, and total limits
• Shared budgets across multiple agents and services
• Guardrails: token limits per run, call count caps, runtime limits, and loop detection
• Multi-provider support: OpenAI, Anthropic, Google (Gemini), and Mistral
• Streaming support with GeneratorExit handling
• YAML configuration for team-wide budget management
• Fail-open design — internal errors never block LLM calls

Alert Channels

• Email notifications via Resend
• Slack webhook integration
• PagerDuty incident creation
• Generic webhook with HMAC signing
• Kill switch callback for emergency shutdowns

Telemetry & API

• OpenTelemetry-based telemetry export
• REST API with JWT and API key authentication
• Redis-backed rate limiting (daily + burst)
• Kafka event pipeline for high-throughput ingestion

Dashboard

• Real-time spend tracking per agent
• Budget configuration and alert rule management
• Run-level detail with event timeline
• Organization-scoped multi-tenant architecture