PetalTrace

PetalTrace is an open-source observability platform for developers working with AI agent workflows. It records prompts and completions, tool calls, token use, costs, and execution timelines, with full capture able to include system prompts, conversation history, and tool definitions. Developers can explore traces, costs, and workflow graphs in a React web interface or use the CLI, HTTP API, and MCP server. Search covers prompt and completion text, while run comparison highlights prompt, output, and cost differences. Captured runs can be replayed using different models or temperatures, or in mocked mode. PetalTrace also accepts standard OpenTelemetry traces from instrumented applications, including those that do not use PetalFlow. PetalFlow integration adds graph topology, node inputs and outputs, and replay-capable snapshots. The documented trace store uses SQLite, and deployment guidance covers building from source or using a release binary to run a local daemon. Authentication is described as future functionality and is disabled by default.

PetalTrace brings trace inspection, cost analysis, run comparison, and replay into one developer-focused platform. Its disabled authentication is worth considering for deployment.

PetalTrace suits developers who need to inspect and compare AI agent workflows. Its CLI, APIs, MCP server, web interface, and self-hosted deployment support different ways to access traces.

Overview

PetalTrace is an open-source observability platform for developers building or operating AI-agent workflows. It is best suited to people who want detailed execution traces and replay tools in a self-managed environment. Its breadth of debugging tools is a strong draw; disabled authentication makes deployment choices important.

Key features

PetalTrace captures prompts and completions, tool calls, token use, costs, and execution timelines. Full prompt capture includes system prompts, message history, and tool definitions, so developers can examine the context behind a model response as well as the response itself. PetalFlow users can choose minimal capture for latency, status, and token counts; standard adds prompts, completions, and tool I/O; full adds graph snapshots and edge data. The trade-off is straightforward: lighter capture records less interaction detail, while full mode preserves more workflow structure.

Run comparison surfaces differences in prompts, outputs, structure, and cost. Replays can use different models or temperatures, and can run live, mocked, or in hybrid mode. That combination makes PetalTrace useful for investigating behavior changes without relying on a single replay setup. Cost tracking by workflow, provider, or model helps locate where spending accumulates; full-text search and real-time SSE feeds support finding past interactions and following active runs.

The CLI, HTTP API, MCP server, and React web UI provide several ways to work with traces. Its MCP server lets AI agents query trace history, inspect prompts, analyze costs, compare runs, and trigger replays, with documented Claude Code configuration. PetalTrace also accepts standard OTLP traces from OpenTelemetry-instrumented applications, including applications that do not use PetalFlow. PetalFlow adds graph topology, node-level inputs and outputs, and replay-capable snapshots.

The documented local store uses SQLite for runs, spans, and LLM interactions, with a default database path of ~/.petaltrace/data.db. Retention defaults to 30 days for runs and 90 days for failed runs, with a maximum period of 365 days. The local-first approach gives operators control over deployment and data, but authentication is disabled by default and described as future functionality. That is a material limitation for any deployment that needs access controls.

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. There is no paid tier to compare, but operators should account for managing their own deployment and retention configuration.

Platforms

PetalTrace supports API, Linux, macOS, self-hosted, web, and Windows. The documented installation paths are building from source with Go or downloading a release binary and running the daemon locally. Its CLI and React web UI cover command-line and browser-based workflows, while the HTTP API and MCP server support programmatic access.

Who it's for

PetalTrace is a strong fit for developers who need to inspect agent prompts, tool activity, costs, and execution flow, especially when they want to replay runs or query trace history through an AI agent. OpenTelemetry support broadens its usefulness beyond PetalFlow-based applications. It is a weaker choice for teams that require authentication in the observability service before deployment.

Pros and cons

• Pros: Captures prompt, tool, cost, and timeline data, with configurable capture depth for PetalFlow workflows.
• Pros: Run comparison and live, mocked, or hybrid replay give developers multiple ways to investigate changes.
• Pros: OpenTelemetry ingestion, MCP tools, and several interfaces make it usable across different workflows and applications.
• Cons: Authentication is disabled by default, a significant constraint for deployments that need access controls.
• Cons: Self-hosting means operators manage the local service and its SQLite-backed data themselves.

Alternatives

SigNoz is worth considering for readers who want a freemium observability option with Android and iOS support as well as API, web, and self-hosted platforms. Its free community plan requires users to manage infrastructure, storage, scaling, upgrades, and backups.

Agenta may suit teams looking for a freemium alternative with a defined free Hobby plan: it allows two team members, 5,000 agent runs per month, 20 evaluations per month, and one-week trace retention.

Traccia is an alternative for readers who prefer a freemium API and web product with a stated event allowance and retention period on its free Hobby plan: 50K events and seven days.

TraceRoot is another freemium option for readers comparing paid monthly plans; its Starter plan is $30 a month and includes unlimited seats, 150k events per month, 30-day retention, and 100 chat runs per month.

VoltOps is an alternative for readers seeking a freemium API, self-hosted, and web option with a free Developer plan limited to one seat, one prompt, one project, and 250 traces per month.

Arize AX offers a freemium alternative whose AX Free plan includes 25k trace spans per month, 1 GB ingestion per month, 15-day retention, and unlimited users and evaluations, with SaaS deployment.

Braintrust may fit readers who want a freemium plan with unlimited users, projects, and datasets; its Starter plan includes 1 GB of processed data, 10,000 scores, and 14-day retention.

Fiddler AI is another freemium option for readers comparing AI observability tools.

AI Agent Observability Tools provides a broader set of alternatives to compare.

Verdict

Choose PetalTrace if you are a developer who wants rich agent traces, flexible replay, and local control in one open-source platform. Its capture, comparison, and MCP capabilities make it particularly compelling for debugging workflows. Look elsewhere if authentication is a deployment requirement you cannot defer.