Observe every AI system.
Out of the box.

Aegara Trace connects to the AI tools your organization already uses. Native adapters for every major provider, automatic detection, and zero-latency instrumentation.

7
Native Providers
30+
Via LangChain
3
Integration Paths
OpenAI
Anthropic
Google Gemini
AWS Bedrock
Cohere
Mistral
Azure OpenAI
LangChain
Groq
Together AI
Perplexity
DeepSeek
OpenAI
Anthropic
Google Gemini
AWS Bedrock
Cohere
Mistral
Azure OpenAI
LangChain
Groq
Together AI
Perplexity
DeepSeek
CrewAI
AutoGen
Fireworks AI
xAI / Grok
Slack AI
GitHub Copilot
Cursor
Microsoft Copilot
ChatGPT
Amazon Titan
Zapier AI
Notion AI
CrewAI
AutoGen
Fireworks AI
xAI / Grok
Slack AI
GitHub Copilot
Cursor
Microsoft Copilot
ChatGPT
Amazon Titan
Zapier AI
Notion AI
Integration Paths

Three ways to connect. Pick what fits.

Whether you have code access, use a framework, or want zero-code setup, Aegara Trace has a path that works.

Recommended

SDK Auto-Wrap

Wrap your AI client in one line. Aegara auto-detects the provider and captures every call with zero latency impact.

Best for: Teams building AI features in their own code.
Supports: OpenAI, Azure OpenAI, Anthropic, Gemini, Bedrock, Cohere, Mistral, and all OpenAI-compatible APIs.
Framework

LangChain Callback

Drop in the Aegara callback handler. Captures LLM calls, tool invocations, chain orchestration, and retriever queries across any provider LangChain supports.

Best for: Agent frameworks (LangChain, CrewAI, AutoGen).
Covers: 30+ providers, full trace linking across multi-step workflows.
Zero Code

API Proxy

Swap the base URL. Aegara proxies your AI calls transparently, captures behavioral metadata, and forwards responses unchanged.

Best for: Quick integration without code changes.
Supports: OpenAI, Anthropic, Gemini, Cohere, Mistral. Full streaming support.
Native Adapters

Built-in support for every major provider

Aegara automatically recognizes which AI provider your team is using. No configuration, no setup, no code changes beyond the initial integration.

OpenAI
GPT-4, GPT-4o, DALL-E, Whisper
Tracks chat, image generation, speech, and transcription actions. Also works with Azure OpenAI, Groq, Together AI, Fireworks, Perplexity, DeepSeek, and xAI.
SDK Proxy 5 endpoints covered
Anthropic
Claude 3.5, Claude 4, Claude Opus
Tracks conversations, tool usage, and token counts. Captures what Claude did and which tools it called, without reading the conversation itself.
SDK Proxy 1 endpoint covered
Google Gemini
Gemini Pro, Gemini Ultra
Tracks text generation, embeddings, and streaming responses. Captures which functions Gemini called and how many tokens it used.
SDK Proxy 3 endpoints covered
AWS Bedrock
Claude, Titan, AI21, Cohere on AWS
Works with any AI model running on AWS, including Claude, Titan, AI21, and Cohere. One integration covers every Bedrock model your team uses.
SDK All Bedrock models
Cohere
Command, Embed, Rerank
Tracks chat, search, and document ranking actions. Captures token usage and tool calls across all Cohere models.
SDK Proxy 3 endpoints covered
Mistral
Mistral Large, Mistral Medium
Tracks chat and embedding actions. Full support for tool calling and streaming. Works the same way as OpenAI integration.
SDK Proxy 2 endpoints covered
Highest Leverage Integration

LangChain Callback Handler

One callback handler covers every provider LangChain supports, now and in the future. Works with LangChain, CrewAI, AutoGen, and any framework that uses the LangChain callback interface. No import of LangChain required.

Every step in an agent workflow is captured as a linked event in a single trace. See exactly what your agent read, processed, called, and returned.

  • T LLM Calls → TRANSFORM events with model name + token counts
  • C Tool Invocations → CALL events with tool name + latency
  • R Chain Orchestration → ROUTE events linking sub-steps
  • D Retriever / RAG → READ events with document counts
  • L Trace Linking → Child runs inherit parent trace IDs
How It Works

From AI call to full visibility

No matter the provider, framework, or integration method, every AI action flows through Aegara and into your dashboard.

OpenAI
GPT-4o, o1, DALL-E
Anthropic
Claude 4, Haiku
Gemini
Pro, Ultra, Flash
Bedrock
Titan, AI21, Cohere
30+ More
via LangChain
SDK Auto-Wrap · LangChain Callback · API Proxy
Aegara Trace
Captures behavioral metadata. Never touches content.
Behavior Events
READ TRANSFORM WRITE CALL STORE ROUTE
Real-Time Dashboard

Live activity feed across all AI systems

Trace Reconstruction

Full action chains for any AI workflow

Audit Trail

Immutable, append-only event log

Behavior Primitives

Every AI action maps to a primitive

Regardless of the provider or model, Aegara normalizes all activity into six fundamental behaviors. This is what makes cross-system analysis possible.

READ

Data Access

RAG document retrieval, audio transcription, email thread reads

TRANSFORM

Data Processing

Chat completions, summarization, embeddings, text generation

WRITE

Output Creation

Image generation, speech synthesis, document drafting

CALL

External Invocation

Tool use, function calling, API invocations via AI agents

STORE

Memory & Persistence

Conversation memory writes, vector store updates, cache saves

ROUTE

Decision & Delegation

Chain orchestration, agent routing, workflow branching

Privacy

What we capture. What we don't.

Aegara monitors behavior, not content. We never read or store your prompts, responses, or API keys. On three of our four integration paths, that content never reaches our servers at all.

Captured

  • ✓ Model name and version
  • ✓ Token counts (input, output, total)
  • ✓ Tool names invoked
  • ✓ Finish reason and latency
  • ✓ Behavior primitive type
  • ✓ Message count (not content)
  • ✓ System prompt presence (not content)

Never Captured

  • ✕ Prompt content or text
  • ✕ Response content or text
  • ✕ API keys or credentials
  • ✕ User personal data
  • ✕ File contents or attachments
  • ✕ Tool arguments or return values
  • ✕ Function call parameters