Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.lumiqtrace.com/llms.txt

Use this file to discover all available pages before exploring further.

​
SDK v1.5.0

Released: April 2026

New

  • wrapBedrock() wrapper for AWS Bedrock converse API
  • wrapMistral() wrapper for Mistral AI
  • wrapGroq() wrapper for Groq
  • wrapLiteLLM() helper for LiteLLM proxy
  • disableConfigSync option to opt out of remote config propagation

Improved

  • wrapGoogle() now captures cachedContentTokenCount and thoughtsTokenCount from Gemini 2.5 responses
  • LumiqtraceCallbackHandler Python handler now supports async LangChain chains natively
  • Flush latency reduced by ~30% via HTTP/2 keep-alive reuse
Bug fixes:
  • Fixed a race condition in the batch engine that could cause duplicate events on process exit under high load
  • Fixed TTFT not being captured for Anthropic streaming responses when stream_options was not set
  • Fixed Python with_agent context manager not propagating to nested sync functions

​
SDK v1.4.2

Released: March 2026 Bug fixes:
  • Fixed cost_usd being computed as NaN for models with zero-cost cached tokens when cached_tokens was undefined
  • Fixed wrapADKRunner not capturing tool spans when the ADK tool result was an empty string
  • Fixed TypeScript types for withLumiqtraceContext not accepting Promise<void> return type

​
SDK v1.4.0

Released: February 2026

New

  • Config propagation — SDK now reads cv from ingest responses and fetches updated config from /v1/sdk/config automatically
  • replaceDefaultRedactKeys option for PII redaction
  • PromptClient.clearCache(name?) — clear cache for a specific prompt or all prompts
  • Google ADK: instrumentADK() convenience function wraps agent + runner in one call

Improved

  • wrapGoogle() now supports the new Gemini 2.5 thoughtsTokenCount field for reasoning token tracking
  • LumiqtraceCallbackHandler for LangChain now automatically propagates userId and sessionId to all nested LLM spans
  • Python SDK: patch_openai() now patches AsyncOpenAI streaming responses for TTFT capture

​
SDK v1.3.0

Released: January 2026 New:
  • wrapOpenRouter() — OpenRouter wrapper using OpenAI SDK interface
  • initOTel() — standalone OpenTelemetry setup function
  • enableOTel and otelEndpoint options in init() for dual-export to OTEL + LumiqTrace
  • Python: @lumiqtrace.trace decorator for sync and async functions
Improved:
  • withAgent now accepts an optional agentId parameter for stable agent identity across runs
  • Batch engine now uses gzip compression by default (was opt-in), reducing payload size by ~75%

​
Platform v2.3.0

Released: April 2026 New features:
  • LumiqPilot config propagation — Pilot can now push SDK config changes (model overrides, sample rate, guardrail toggles) to running instances in real time
  • Incident auto-remediation — Scale plan: configure remediation rules that Pilot applies automatically when an incident opens
  • Simulation batch runs — compare multiple prompt versions in parallel against the same dataset
  • Session view — new dashboard page grouping LLM calls by session_id
Improvements:
  • Anomaly detection now uses 14-day rolling baseline (was 7 days) for fewer false positives
  • Cost Optimizer refresh cadence: Scale plan moved from daily to near-real-time (sub-1-hour)
  • Trace flame graph now renders guardrail spans inline with their parent LLM span
Bug fixes:
  • Fixed alert history table not loading for projects with >10K alert events
  • Fixed POST /v1/otel accepting spans without gen_ai.* attributes but dropping them silently — they now appear in traces as generic spans

​
Platform v2.2.0

Released: February 2026 New features:
  • Evaluations dashboard — LLM-as-judge evaluator definitions, auto-run mode, trend charts, per-trace scores
  • Simulations — dataset management, scenario configuration, batch runs with side-by-side comparison
  • Guardrails dashboard — execution history, per-policy trigger frequency, AI-powered guardrail types (toxicity, prompt injection, custom judge)
  • Prompt management dashboard — version history, label management, protected labels, dependency graph
Improvements:
  • NLQ result summaries now use streaming for sub-second time-to-first-word
  • Dashboard overview page loads 2× faster via improved ClickHouse query plans

​
SDK v1.2.0

Released: December 2025 New:
  • GuardrailBlockedError class with phase, results properties
  • Guardrails support on wrapAnthropic, wrapGoogle, and LumiqtraceCallbackHandler
  • PromptClient — fetch, compile, create, and manage versioned prompts at runtime
  • Python: PromptClient with same API as TypeScript version
Breaking change:
  • wrapOpenAI(client, options) second argument changed from WrapOptions | undefined to a full options object. If you were passing undefined, no change needed.

​
SDK v1.0.0

Released: October 2025 Initial public release.
  • initLumiqtrace() / lumiqtrace.init() singleton initialization
  • wrapOpenAI(), wrapAnthropic(), wrapGoogle() provider wrappers
  • withLumiqtraceContext() — async context propagation
  • withAgent() — multi-agent tracing with plan, tool, and handoff spans
  • startSpan() — manual span API with eval score attachment
  • LumiqtraceCallbackHandler for LangChain (TypeScript + Python)
  • wrapADKAgent() and wrapADKRunner() for Google ADK
  • PII redaction with configurable key list
  • Batch engine with gzip NDJSON transport
  • Python SDK parity: all TypeScript features available in Python