PetalTrace

Overview

PetalTrace is an open-source observability tool for developers who need to inspect and debug AI-agent workflows. It is strongest for teams that want detailed traces, run comparison, and replay in a self-managed setup; the main caution is that authentication is disabled by default.

A CLI, HTTP API, MCP server, and web interface put trace data within reach of both developers and AI agents. SQLite storage and local deployment suit teams that want control over their own trace store, but this is not the obvious choice for a team seeking a managed service with authentication ready to use.

Key features

PetalTrace records prompts and completions, tool calls, token usage, costs, and execution timelines. Full prompt capture includes system prompts, message history, and tool definitions, so developers can investigate what the model received as well as what it returned. PetalFlow users can choose minimal capture for latency, status, and token counts; standard capture adds prompts, completions, and tool I/O; full capture adds graph snapshots and edge data. That flexibility helps teams balance workflow context against the amount of interaction content they retain.

Run comparison identifies prompt, output, structural, and cost differences, while replay can use different models or temperatures and can run live, mocked, or hybrid. Together these tools make PetalTrace more useful for investigating a change than a trace viewer limited to post-run inspection. Costs can be tracked by workflow, provider, or model, and full-text search across prompts and completions plus real-time SSE feeds help with finding past activity and following active runs.

The MCP server lets agents query trace history, inspect prompts, analyze costs, compare runs, and trigger replays; the documentation includes Claude Code configuration. OpenTelemetry support also accepts standard OTLP traces from instrumented applications that do not use PetalFlow, broadening its reach beyond that integration. PetalFlow adds graph topology, node-level inputs and outputs, and replay-capable snapshots.

The documented store uses SQLite, with a default database at ~/.petaltrace/data.db. Retention defaults keep ordinary runs for 30 days and failed runs for 90 days, with a maximum of 365 days. Those defaults make retention a deployment setting to plan around, rather than an unlimited archive.

Pricing

PetalTrace is open source under the MIT license, and the PetalTrace plan is 0.00 USD per free. The public repository has no stated commercial pricing or usage limits. There is no paid tier to compare: the trade-off is between using a free, self-managed tool and accepting responsibility for its deployment and storage.

Installation is documented through building from source with Go or downloading a release binary and running the daemon locally. The React web UI and CLI provide ways to explore traces, costs, and workflow graphs, but the local deployment model and disabled-by-default authentication deserve careful consideration before exposing the service beyond a trusted environment.

Platforms

PetalTrace supports API, Linux, macOS, self-hosted, web, and Windows. Its CLI and HTTP API suit programmatic workflows, while the web UI provides a visual route into traces and workflow graphs. The MCP server adds agent-facing access; SQLite keeps the documented trace store local to the deployment.

Who it's for

PetalTrace is best suited to developers working on AI-agent workflows who need to understand prompts, tool activity, costs, and execution structure, then compare or replay runs while debugging. It is a particularly strong fit for PetalFlow users seeking graph context and for teams that want agents themselves to query trace history through MCP. Teams that need authentication enabled as a default deployment feature should look elsewhere or account for that limitation before deployment.

Pros and cons

• Pros: Captures prompts, tool calls, costs, and timelines, with configurable capture depth for PetalFlow users.
• Pros: Combines run comparison, replay, cost analysis, search, and active-run streaming in one developer-facing toolset.
• Pros: MCP access and standard OTLP support extend trace analysis to AI agents and OpenTelemetry-instrumented applications beyond PetalFlow.
• Cons: Authentication is disabled by default, a significant concern for deployments that are not kept within a trusted boundary.
• Cons: Local SQLite storage and documented local daemon deployment make operating and retaining trace data the team's responsibility.

Alternatives

OpenLIT is a reasonable alternative for teams who want an open-source option with unlimited self-hosted usage, users, projects, and environments, plus community support.

W&B Weave may suit teams that prefer a freemium plan with a stated monthly ingestion and storage allowance, alongside evaluations, tracing, and scorers.

Opik is worth considering if a team wants a free cloud plan or the option to download and run its open-source observability and evaluation features locally.

SigNoz is an alternative for teams seeking broader platform coverage; its community self-hosted plan leaves infrastructure, storage, scaling, upgrades, and backups to the user.

Agenta offers a free Hobby plan with stated team, run, evaluation, and trace-retention limits, which can help a small team judge a bounded allowance against its needs.

Arize AX may be a better fit for teams that prefer a SaaS free tier with stated trace-span, ingestion, issue, and retention limits.

Braintrust offers a free Starter plan with stated processed-data, score, and retention limits alongside unlimited users, projects, and datasets.

Galileo is another option for teams willing to consider a paid plan priced at 100.00 USD per month, billed yearly, with a stated monthly trace allowance and standard RBAC.

See also AI Agent Observability Tools for more options in this category.

Verdict

Choose PetalTrace if your team wants open-source, self-hosted observability that connects detailed agent traces to comparison and replay, especially with PetalFlow or an MCP-driven workflow. Its strongest reason to look elsewhere is operational: authentication is disabled by default, and the local deployment puts storage and retention in your hands.