Langfuse Integration

Overview

Import traces from your Langfuse project. Converra extracts the system prompt and a readable transcript so you can optimize based on real production behavior.

What Gets Imported

Converra imports traces from Langfuse and classifies them by type:

• Conversational traces (2+ user messages): Chatbots, support agents, copilots
• Single-shot traces (1 user message): Summarization, classification, extraction
• Prompt: The system prompt from each GENERATION observation
• Transcript: A flattened, readable conversation built from LLM inputs/outputs
• Basic LLM params: model + sampling params when available

Note

Tool calls are captured and displayed in run details. Full tool schema preservation for replay is on the roadmap.

What Gets Skipped

Converra automatically filters out traces that can't be optimized:

Trace Type	Example Use Case	Action
No system prompt	Embeddings, simple completions	Skipped
No GENERATION observations	Spans without LLM calls	Skipped
Already imported	Duplicate trace + observation ID	Skipped

How It Works

1. Connect - Paste your Langfuse public key and secret key
2. Select host - Choose the EU or US cloud region, or enter your self-hosted URL under Advanced: Self-hosted
3. Import - Click "Sync Now" to fetch recent traces (or enable auto-sync)
4. Review prompts - Prompts are deduped by content hash and shown in Converra for optimization

Continuous Sync

Enable auto-sync to keep your Converra data fresh with new traces from Langfuse automatically.

Sync Settings

Setting	Options	Description
Auto-sync	On / Off	Enable automatic syncing on a schedule
Frequency	Hourly, Every 6h, Every 12h, Daily	How often to sync new traces
Preferred time	0-23 UTC	For daily syncs, which hour to run

How Auto-Sync Works

1. Watermark-based - Only fetches traces newer than the last sync
2. Incremental - Skips already-imported traces automatically
3. Self-healing - Pauses after 5 consecutive failures (check credentials)
4. Overlap window - 5-minute lookback to catch late-arriving traces

Manual Sync

Click Sync Now at any time to trigger an immediate sync. This resets the watermark to fetch traces from the last 24 hours.

Key Differences from LangSmith

Aspect	LangSmith	Langfuse
Authentication	API Key	Public Key + Secret Key (Basic Auth)
Project Selection	Required (select from list)	Not needed (API keys are project-scoped)
Region	N/A	EU or US
Single-shot traces	Skipped	Imported (classified separately)

Compatibility

Converra auto-detects your tracing architecture and parses common LLM formats.

Supported Architectures

Architecture	Description	Status
Simple chatbot	Single GENERATION observation	✅ Supported
Multi-agent (per-trace)	Multiple GENERATION observations in a trace	✅ Supported

Supported LLM Formats

Converra parses these input/output formats automatically:

Inputs:

• OpenAI: input.messages[] with role/content
• Anthropic: input.system + input.messages

Outputs:

• OpenAI: output.choices[0].message.content
• Anthropic: output.content[] text blocks
• Direct: output.text, output.content

Agent Role Detection

For multi-agent systems, Converra classifies agents by name:

• Routers: Names containing "orchestrator", "router", "coordinator", "dispatcher"
• Responders: All other agents (treated as conversational)

Have a setup that's not working? Contact us and we'll help.

Multi-Agent Support

Langfuse projects often contain multi-agent traces (Router → SupportBot → RefundHandler).

Converra discovers all GENERATION observations in a trace and groups them into agent systems:

• Agent Discovery - See which prompts appear together in traces
• Path Visualization - View the actual execution paths
• Per-Prompt Scores - Each prompt in the system gets its own performance metrics
• Prompt Type Classification - Conversational prompts vs single-shot

Navigate to the Agents page to see discovered agent systems and their performance.

Getting Started

1. Create an account to enable the connector
2. Go to Integrations in the Converra app
3. Click Connect Langfuse
4. Enter your Langfuse API credentials:
   • Public Key: pk-lf-...
   • Secret Key: sk-lf-...
   • Host: EU (cloud.langfuse.com), US (us.cloud.langfuse.com), or a custom self-hosted URL under Advanced: Self-hosted
5. Click Connect and then Import Runs

Finding Your API Keys

1. Log into Langfuse Cloud (EU) or US Cloud
2. Go to Project Settings → API Keys
3. Create a new key pair or use an existing one
4. Copy both the Public Key and Secret Key

TIP

Langfuse API keys are project-scoped, so you don't need to select a project after connecting. The keys determine which project's data is imported.

FAQ

Do I need to change my Langfuse instrumentation?

No. If you already trace runs in Langfuse, Converra imports what you have. You can optionally add more detail for better replay fidelity.

How is this different from Langfuse's prompt management?

Langfuse provides excellent prompt versioning and A/B testing. Converra is complementary: we turn your traces into measurable, simulation-tested improvements you can ship with confidence. Use Langfuse for versioning, Converra for optimization.

Can I import from self-hosted Langfuse?

Yes. Open Advanced: Self-hosted while connecting Langfuse and enter the base URL of your instance. Converra uses that URL instead of the selected Langfuse Cloud region.

What about continuous sync?

Continuous sync is available! Enable Auto-sync in the integration settings to automatically import new traces on a schedule. See Continuous Sync for configuration options.