PetalTrace

Overview

PetalTrace is an open-source observability tool for developers building AI-agent workflows. It suits developers who want detailed trace inspection and replay in a self-managed environment; disabled authentication makes it a poor fit for an exposed shared service without additional safeguards.

Its combination of prompt-level tracing, run comparison, and replay offers a focused way to investigate agent behavior. SQLite-backed local storage and several interfaces give developers flexibility, but the local deployment model puts operational responsibility on the user.

Key features

PetalTrace records prompts and completions, tool calls, token use, costs, and execution timelines. Full prompt capture preserves system prompts, message history, and tool definitions, which helps diagnose what context an agent received—not just what it produced. PetalFlow users can select minimal capture for latency, status, and token counts; standard capture adds prompts, completions, and tool I/O; full capture also records graph snapshots and edge data. Lighter capture reduces recorded detail, while full capture gives more workflow context.

Run comparison surfaces differences in prompts, outputs, structure, and cost. Replays can use different models or temperatures and run live, mocked, or hybrid, giving developers ways to investigate changes without relying solely on a live execution. Costs can be tracked by workflow, provider, or model. Full-text search across prompts and completions and real-time SSE feeds for active runs add useful ways to follow and find activity.

The CLI, HTTP API, MCP server, and React web UI cover both interactive inspection and integration into developer workflows. The MCP server lets agents query trace history, inspect prompts, analyze costs, compare runs, and trigger replays; the documentation includes Claude Code configuration. PetalTrace also accepts standard OTLP traces from OpenTelemetry-instrumented applications, including ones that do not use PetalFlow, broadening its use beyond that framework.

The documented store uses SQLite, with a default database at ~/.petaltrace/data.db. Retention defaults keep runs for 30 days and failed runs for 90 days, with a maximum period of 365 days. That local, file-based approach favors users who want to manage their own trace data, but also leaves storage and deployment operations in their hands. Authentication is disabled by default and described as future functionality, a significant caveat for shared or network-accessible deployments.

Pricing

PetalTrace is open source under the MIT license, and its PetalTrace plan costs nothing. The public repository has no stated commercial pricing or usage limits. That makes it a straightforward option for developers who can deploy and operate it themselves; the tradeoff is self-management rather than a priced hosted service with stated usage allowances.

Platforms

PetalTrace supports API, Linux, macOS, Windows, web, and self-hosted use. The getting-started guide covers building from source with Go, and release binaries are also available. The documented local daemon and React web UI suit developers comfortable managing a local or self-hosted installation.

Who it's for

PetalTrace is best suited to developers diagnosing agent workflows who need prompt and tool context, cost visibility, run comparison, and replay. PetalFlow users can choose how much execution detail to retain, while OpenTelemetry users can send standard OTLP traces without adopting PetalFlow. It is less suited to teams that require authentication in the product for shared deployment or prefer not to manage local storage and operations themselves.

Pros and cons

Pros: Captures prompts, tool calls, costs, and timelines, then connects comparison and replay in the same workflow—useful for investigating both behavior and execution changes.

Pros: OpenTelemetry ingestion extends trace collection to instrumented applications beyond PetalFlow.

Pros: The MIT-licensed free plan has no stated usage limits, and offers several interfaces, including MCP tools for agent-driven trace analysis.

Cons: Authentication is disabled by default, limiting its suitability for shared or exposed deployments without other safeguards.

Cons: SQLite storage and local deployment put data retention and operational management on the user.

Alternatives

AI Agent Observability Tools is a broader category guide for readers comparing options across the field.

W&B Weave is worth considering for its free tier, which includes 1 GB/mo Weave data ingestion, 5 GB/mo storage, and AI application evaluations, tracing, and scorers.

SigNoz offers a self-hosted community plan for users prepared to manage infrastructure, storage, scaling, upgrades, and backups themselves.

Opik is an alternative for developers who want open-source observability and evaluation features they can download and run locally.

OpenLIT may suit users seeking an OSS plan with unlimited self-hosted usage, users, projects, and environments.

Traccia offers a free Hobby plan with 50K events and 7 days of retention, or an Observe plan at $99 a month with 500K events.

VoltOps has a Developer plan at free with limits including one seat, one prompt, one project, and 250 traces per month.

Helicone describes itself as open source and may suit readers looking for another community-oriented option.

Maxim AI offers a Developer plan with up to three seats, one workspace, up to 10k logs per month, 3-day data retention, and email support.

Verdict

Choose PetalTrace if you are an AI-agent developer who wants open-source, replay-oriented observability and can manage a local or self-hosted service. Its strongest case is the combination of rich trace capture, comparison, replay, and OpenTelemetry ingestion at no stated usage limit. Look elsewhere if built-in authentication is a requirement for your deployment.