AtlasBurn
Paid
AI-native teams tracking variable inference costs
Orchid
FreeOpen SourceSelf-Hosted
When an agent fails on step seven of a twelve-step run, you are back to print statements and grep — unless you captured the whole thing before it broke. Orchid is a zero-instrumentation proxy that records every LLM call and tool invocation your agent makes, then lets you replay and inspect each step offline.
Similar tools
AtlasBurnPaidAtlasBurn monitors token usage and cost distribution across OpenAI, Anthropic, Gemini, and AWS…
AgentmetryFreeAgentmetry is an open-source, local-only session recorder for AI coding agents — Cursor, Claude…
Selfship.aiPaidThe observe-diagnose-fix-verify cycle runs on live traffic without you assigning tasks or…
SemelbaseFreemiumThe vendor positions Semelbase as a ledger and routing layer that sits between your agents and…
Get This Tool
License: Apache-2.0 Any use incl. commercial
Local-run terms: Run the open-source proxy container and SDKs locally; full source available under Apache-2.0.
Summary
Orchid sits between your agent and any API it talks to, capturing traffic into a local SQLite file — no cloud account, no SDK changes, no telemetry leaving your machine. The built-in web UI lets you step through a completed run, inspect every prompt, response, token count, and cost. The proxy also runs a built-in MCP server, so your IDE assistant in Cursor, VS Code, or Claude Code can query recorded traffic directly. Replay is deterministic and costs nothing in API fees. The ceiling appears when your team needs cross-service aggregation or production alerting — this tool is a local debugger, not an observability platform.
Bottom line: Pick Orchid when you are debugging a failing LangGraph agent locally and need step-by-step replay without a cloud dependency — reach for something else when you need dashboards, alerting, or traces aggregated across a fleet of production agents.
Find your fit
Before you choose Orchid
More options in LLM Observability. Compare pricing, platforms, and features side by side.
Agentmetry
Free · Open source
SOC and DevSecOps teams
Pros
- Zero-instrumentation proxy capture, so you get full LLM and tool-call traces without touching your agent's source code — no retrofit required when a bug surfaces in a framework you do not control.
- Deterministic offline replay from recorded SQLite files, which means you reproduce a specific failure run as many times as you need without burning API credits on each attempt.
- Built-in MCP server lets IDE assistants query recorded traffic directly, so instead of manually hunting through log output you ask your coding assistant what happened on step six.
- Local SQLite storage with no cloud dependency, so teams under data-sensitivity constraints get full trace visibility without any traffic leaving the machine.
- Apache-2.0 open-source license, which means you can inspect, fork, and modify the proxy to fit your stack — no vendor lock-in on your debugging infrastructure.
Cons
- Orchid captures traffic on a single local machine for a single agent run; there is no aggregation across parallel runs or distributed agent instances. Teams running agents across multiple services or needing a unified view of production traffic hit this wall immediately and move to a dedicated observability platform.
- No alerting, no anomaly detection, no dashboards shared across a team. When the use case shifts from 'reproduce this specific failure' to 'monitor agent health in production,' Orchid has nothing to offer — teams at that stage switch to platforms built for production observability.
- The GitHub repository shows 3 stars and 47 commits at the time of curation, with no public issue activity. Early-stage projects at this scale carry real risk: breaking changes ship without deprecation windows, documentation gaps appear in edge cases, and community support for non-obvious integration problems is thin.
About
- Platforms
- Docker, local
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- June 25, 2026
Best For
Who it's for
- Developers building and debugging multi-step AI agents
- Teams needing local, private observability without cloud services
- LangGraph and similar framework users
What it does well
- Debugging AI agent failures by inspecting full traces
- Replaying agent runs deterministically offline
- Querying recorded traffic directly from IDE assistants via MCP
- Capturing LLM and tool calls with minimal changes
Integrations
Frequently Asked Questions
- Is Orchid free?
- Orchid offers a free option. Hosting, model usage, or optional services may have separate costs; check the pricing details and official terms.
- Is Orchid open source?
- Yes. Orchid is open source.
- Can I self-host Orchid?
- Yes. Orchid supports self-hosting on your own infrastructure.
- What platforms does Orchid support?
- Orchid is available on: Docker, local.
Curated lists that include this category
Debugging a multi-step agent without a trace is an exercise in working backward from a broken final state through layers of framework abstraction you did not write. Orchid solves this by running as a local proxy: your agent’s network traffic — LLM completions, tool calls, any other API request — flows through Orchid and gets written to a local SQLite database. No code changes to your agent are required. After a run, you open the built-in web UI and step through the exact sequence of prompts, responses, token counts, and costs, in order, from the moment the run started.
The differentiating feature is the built-in MCP server. Because the proxy captures structured records of every call, an AI coding assistant running in Cursor, VS Code, or Claude Code can query those records directly via MCP. The vendor describes this as giving your coding agent the ability to debug your AI app — meaning when a failure is buried behind three layers of framework abstraction, you do not have to manually correlate log lines; your IDE assistant can query the trace and answer questions about it.
Orchid fits teams doing active development and debugging of multi-step agents who want local, private observability with zero cloud dependencies. The Apache-2.0 license and self-hosted SQLite storage mean no data leaves the machine. Where this breaks: there is no production alerting, no aggregation across multiple agent instances, and no hosted dashboard for sharing traces across a team. Teams that graduate from local debugging to monitoring agents in production will need a separate observability layer — Orchid does not replace that.
The repo documents an SDK and MCP integration path alongside the proxy itself. LangGraph is called out specifically in the documentation as a target framework, though the proxy’s network-layer approach means it captures traffic regardless of the framework above it.
Contribute
- Work on Orchid? Claim this listing.
- Embed a badge for your README or site.
Sign in to rate this tool or suggest an edit.