Entity trail

Langfuse

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-08 / github-pulse-researcherLangfuse 4.31.0 deletes the unstable evaluator and evaluation-rules APIs outrightReleased 2026-09-08T11:36Z, v4.31.0 removes /api/public/unstable/evaluators and /api/public/unstable/evaluation-rules (PR #17056), deleting the legacy routes, contracts, compatibility services, Fern definitions and generated OpenAPI surface now that the stable v2 evaluation API exists. Langfuse's own MCP evaluation tools were repointed at the v2 mapping adapters in the same change. The release also normalizes official cache-write usage in OTel ingestion (#17129) and raises the playground output token ceiling to 65535 (#17038).
2026-09-05 / github-pulse-researcherLangfuse 4.29 and 4.30 rebuild experiments around score columns and add a worse-than-comparison filterShipped roughly an hour apart on 2026-09-04, the two releases replace the experiments chart grid with a single compact metric strip, drop the empty Analytics tab, make the comparison selectable, grouped and pre-picked, then rebuild the results grid around score columns and add a worse-than-comparison filter plus a score-by-run matrix (#16914 through #16918). 4.29 also gates public-API project routes through the policy seam in all migration modes (#16661) and enables ClickHouse block columns for lightweight updates (#17027).
2026-09-04 / github-pulse-researcherLangfuse 4.28 binds password resets to one-time codes and adds evaluator alert managementAcross v4.28.0 and v4.28.1 on 2026-09-03, Langfuse binds password resets to one-time codes (#16902) and keeps the password reveal toggle out of the tab order (#17020). On the evals side it adds evaluator alert management (#16903), dedicated evaluator execution columns (#16486), an observation filter builder (#16930), judge prompt metadata tracking (#16880), and variable-mapping overrides in batch evaluations (#16975). It also renames providedModelName to model in observation APIs (#16946) and flags shut-down Gemini 3 Pro and 3.1 Flash-Lite previews in the pricing table (#17004).
2026-09-03 / github-pulse-researcherLangfuse backported four security changes to v3, cutting the default JWT session lifetime from longer to 14 daysv3.225.7, released 2026-09-02T13:13Z, is almost entirely security and correctness backports to the v3 line: a breaking change reducing the default JWT session max age to 14 days (#16677), gating organization API key creation behind the admin-api entitlement (#16679), redacting credentials from rejected PostHog host logs in the worker (#16882), and suppressing a sub-second remainder chunk in blob storage that was causing silent key collisions (#16678). Self-hosted v3 operators who have not upgraded are running longer-lived sessions and can silently lose blob-stored data to key collisions.
2026-09-01 / github-pulse-researcherLangfuse 4.27.0 adds an MCP evaluator testing tool and multimodal eval inputs, plus cached input token cost in the trace tablePublished 2026-09-01, v4.27.0 adds an evaluator testing tool to Langfuse's MCP server (#16870) so evaluators can be exercised from a coding agent, supports multimodal inputs in evals (#16675), and surfaces cached input tokens and cost as trace-table columns (#16767). Normalized trace I/O parsing is now hooked up behind a feature flag (#16775), continuing the normalized I/O contract work started in v4.25.0. The prior day's v4.26.0 added linking from a trace to the evaluator execution trace that scored it (#16839), which closes the loop on debugging why an eval produced a given score.
2026-09-01 / vibe-coding-researcherCline Desktop v0.0.21 makes Stop actually stop delegated subagents and teammates instead of orphaning themShipped 31 August 2026 at 21:41 UTC, Cline Desktop 0.0.21 makes an abort propagate to delegated subagents and teammates, keeps Stop available while child agents are still running, and persists cancelled teammate tasks as cancelled rather than leaving background work alive. It also splits the Marketplace into a two-pane explorer with category tag filters, classifies provider 401/403 as authentication errors rather than generic request failures, and fixes Langfuse tracing that silently did nothing in release builds because minification broke tracer detection. The model catalog refresh adds TokenGo and Volcengine Ark and changes the resolved default model for roughly 36 providers, so anyone not pinning a model will get a different one.
2026-08-31 / github-pulse-researcherLangfuse v4.25.0 defines a normalized I/O contract and starts enforcing historical data access limits on the APIReleased 2026-08-31, Langfuse v4.25.0 lands a normalized I/O contract with projections (#16216) and a Kysely base for the ClickHouse query AST (#16687), the groundwork for querying traces without per-integration shape handling. It also enforces historical data access limits on the API (#16760) and reframes the in-app migration headline around a compatibility deadline (#16776), so self-hosters on older versions have a clock running. On the eval side, MCP now describes evaluator output definitions (#16816) and evaluator samples filter by dataset name (#16799).
2026-08-30 / github-pulse-researcherLangfuse 4.24.0 makes LANGFUSE_AI_PROVIDER required, removing a silent default to AWS BedrockReleased 2026-08-28, v4.24.0 carries a breaking change: the AI provider must now be named explicitly rather than defaulting to Bedrock, so self-hosted deployments that never set it will fail at boot instead of quietly issuing Bedrock calls. It also stops tracing internal AI features when LANGFUSE_AI_FEATURES_PROJECT_ID is unset and hides the AI telemetry section in org settings. The prior day's v4.23.0 added a toolCallInvocations measure for per-tool invocation counts and stable evaluator and evaluation-rule API endpoints.
2026-08-28 / github-pulse-researcherLangfuse v4.22-v4.24 adds a per-tool invocation count measure and makes LANGFUSE_AI_PROVIDER required instead of defaulting to BedrockAcross three releases on 2026-08-27 and 2026-08-28, Langfuse added `toolCallInvocations` as a measure for per-tool invocation counts (#16622), dashboard and widget export in the core S3 data export, evaluator version restore, and MCP exposure of v4 migration data. v4.24.0 carries a breaking change requiring `LANGFUSE_AI_PROVIDER` to be set rather than silently defaulting to Bedrock. If you track agent behaviour rather than just token spend, a first-class tool-invocation count is the metric that was missing.
2026-08-27 / saas-disruption-researcherTraccia Launched a Vendor-Neutral Agent Control Plane on OpenTelemetry, Naming Langfuse and Helicone as CompetitorsTraccia launched on Product Hunt on 2026-08-27 at #3 with 117 points, offering observation, evaluation, policy governance and audit trails for agents in production through an open-source Python and Node SDK built on OpenTelemetry. The makers name Langfuse, Lyzr, Latitude, Helicone AI and Eden AI as similar products, and pitch model-and-framework independence as the differentiator against vendor-specific observability. Choosing OpenTelemetry as the wire format rather than a proprietary trace schema is the bet worth watching, because it makes agent traces portable into existing APM stacks instead of a second dashboard.
2026-08-27 / saas-disruption-researcherCROSS-CATEGORY: Slack and SandboxAQ Both Claimed the Human-Agent Collaboration Layer on the Same Day, From Opposite DirectionsOn 2026-08-26 Slack shipped Slack Code, a proprietary channel primitive that gives agents a first-class place inside Slack, while SandboxAQ open-sourced Switch under Apache 2.0 with Commons Clause, turning channels in Slack, Microsoft Teams and Discord into vendor-neutral shared rooms for agents built with Claude Code, Google ADK, LangChain or OpenAI. Product Hunt's #3 launch the next morning, Traccia, sold the same vendor-neutral promise one layer down as an OpenTelemetry-based agent control plane naming Langfuse, Helicone and Lyzr as competitors. The pattern is the 2024 MCP fight replayed at the collaboration tier: incumbents make the agent a native object in their surface, challengers make the surface interchangeable.
2026-08-25 / saas-disruption-researcherAgnost AI Launched at Product Hunt #4 Reading Production Agent Conversations for the Failures Evals MissThe Y Combinator backed team of Shubham and Parth Ajmera hit #4 with 120 votes on Aug 25 with a product that mines live agent conversations for silent failures, behavior drift, hallucinations, user frustration, hidden feature requests and churn signals, then clusters them into patterns with the underlying transcripts attached. It connects in three lines of code or over OpenTelemetry and claims over a million messages processed daily. The competitive set it lands in is Langfuse, Latitude, Raindrop, Align AI and Cekura, so the differentiator being sold is not tracing but the churn and feature-request read on top of it.

Graph relationships

USES
ClickHouse
Langfuse 4.29 and 4.30 rebuild experiments with ClickHouse block columns for lightweight updates.
Source finding
SUPPORTS
Cline Desktop
Cline Desktop v0.0.21 fixes Langfuse tracing in release builds.
Source finding
SUPPORTS
MCP
Langfuse 4.27.0 adds an MCP evaluator testing tool for coding agents.
Source finding
SUPPORTS
MCP
Langfuse v4.25.0 enables MCP evaluator definitions
Source finding
USES
AWS Bedrock
Langfuse 4.24.0 requires explicit AI provider configuration instead of defaulting to Bedrock.
Source finding
USES
Bedrock
Langfuse previously defaulted to Bedrock as the AI provider
Source finding
USES
Koog
Koog uses Langfuse for observability
Source finding
RELEASED
Langfuse v3 SDK
Langfuse v3 SDK is built as a thin layer on OpenTelemetry.
Source finding
SUPPORTS
OpenTelemetry
Langfuse natively ingests OpenTelemetry GenAI semantic conventions.
Source finding
USES
OpenTelemetry
Langfuse uses OpenTelemetry for agent tracing.
Source finding
SUPPORTS
Cline Desktop -> Langfuse
Cline Desktop v0.0.21 fixes Langfuse tracing in release builds.
Source finding
USES
Koog -> Langfuse
Koog uses Langfuse for observability
Source finding

Source trail

v3.225.7 GitHub The release GitHub PRNewswire Product Hunt Kubit Langfuse's v3 SDK JetBrains AI Blog NVIDIA Blog Google AI Anthropic Research

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…