Traccia
traccia.ai
- Free plan
- Web
- API
What Traccia is
Traccia is an AI agent control plane for observing agent activity, evaluating outputs, applying governance, and enforcing policies at runtime. Its SDK captures runs, sessions, tool calls, prompts, and errors from Python and TypeScript. Spend limits, model allowlists, and loop caps can stop a call, while detective rules log issues after a run. Evaluation tools bring together prompts, datasets, scorers, and experiments to assess outputs before promotion. Integrations listed in its documentation include LangChain, Claude Code, CrewAI, OpenAI Agents SDK, and Gemini SDK. Traccia says it works with any LLM; it automatically tracks tokens and costs for OpenAI and Anthropic, and supports OpenTelemetry tracing for HTTP-based providers. Its open-source SDK can export traces to compatible backends without an account. The free Hobby plan includes 50K events and seven days of retention; paid plans start at $99/month. Traccia says SOC 2 Type II certification is in progress and does not offer a signed Business Associate Agreement.
Who it suits
It suits teams building AI agents that need trace capture, output evaluation, or runtime controls. Teams handling protected health information should note Traccia's stated BAA limitation.
Good to know, and what to watch
- SDK captures runs, sessions, tool calls, prompts, and errors.
- Runtime controls include spend limits and loop caps.
- Evaluation tools include datasets, scorers, and experiments.
- Open-source SDK can export traces without an account.
- Optional pattern-based PII masking runs before spans leave the process.
- Hobby plan retains data for seven days.
- SOC 2 Type II certification is in progress.
- No signed Business Associate Agreement is available.
- ML-based PII detection is planned, not listed as available.
Plans
Every plan and priceEnterprise
Hobby
Observe
Govern
Free plan
Getting started
- Open the Traccia website or web interface
- Choose the free Hobby plan or a paid plan starting at $99 per month
- Install the Python or TypeScript SDK
- Connect an agent framework integration such as LangChain, CrewAI, or OpenAI Agents SDK
- Configure tracing, policy controls, evaluations, and optional PII masking
- Export traces to a compatible OpenTelemetry backend when using the open-source SDK
Limits to check
The free Hobby plan includes 50K events and seven days of retention. Observe costs $99 per month for 500K events and 30 days of retention, with overage charged at $12 per 100K events.
The Club's review
Traccia is an AI-agent control plane for teams that need observability tied to evaluation and runtime guardrails. It fits developers and platform owners working in Python or TypeScript who want to inspect complete agent workflows and stop selected calls while they run. The free entry tier is useful for small experiments, but retention, event ceilings and compliance constraints become important as usage grows.
Overview
Traccia brings tracing, governance, evaluation and policy enforcement into one hosted service, with a hybrid deployment option. Its SDK records runs, sessions, tool calls, prompts and errors rather than only individual model requests. That broader view is valuable when an agent's failures arise from orchestration or tool use, not simply from the model response.
The service is also designed to intervene. Spend limits, model allowlists and loop caps can terminate a call at runtime, whereas detective rules record what happened after execution. This separation is practical: preventative controls can contain runaway or disallowed behavior, while detective rules support review without blocking the workflow.
EU-region processing can be configured for production, with EU-only storage confirmed contractually during onboarding. Granular role-based access control is available. Traccia says SOC 2 Type II certification is in progress, and its HIPAA module does not by itself establish compliance; organizations requiring a signed BAA should look elsewhere.
Key features
The Python and TypeScript SDK covers agent runs, sessions, tools, prompts and errors. Documented integrations include LangChain, Claude Code, CrewAI, OpenAI Agents SDK and Gemini SDK, giving teams using those frameworks a direct starting point. Traccia says it works with any LLM. Automatic token and cost tracking is provided for OpenAI and Anthropic, while HTTP-based providers can be traced with OpenTelemetry.
Evaluation uses prompts, datasets, scorers and experiments to assess outputs before promotion. That makes the product relevant to teams establishing release gates for agent behavior, not just diagnosing incidents in production. Session replay and prompt and tool tracing add context when reconstructing a run.
The open-source SDK can export traces to Jaeger, Grafana Tempo, Zipkin or another OpenTelemetry-compatible backend without a Traccia account. This is a strong option for organizations that want local control or an existing telemetry stack, although using those destinations does not provide Traccia's hosted policy and evaluation workflow. Optional pattern-based PII masking can run before spans leave the user's process; machine-learning PII detection is planned rather than present.
Pricing
| Plan | Price | Included usage and retention |
|---|---|---|
| Hobby | Free | 50,000 events; 7 days |
| Observe | $99/month | 500,000 events; $12 per 100,000 overage; 30 days |
| Govern | $299/month | 2 million events; $8 per 100,000 overage; 90 days |
| Scale | $799/month | 10 million events; $5 per 100,000 overage; 1 year |
| Enterprise | Custom pricing | Volume events, custom overage; 10 years |
Hobby is a sensible way to instrument a pilot, but seven-day retention and a 50,000-event cap limit historical analysis. Observe suits a small production deployment and is the first paid step, while Govern is aimed at teams needing more volume and three months of history. Scale's lower overage rate and one-year retention fit substantial workloads; Enterprise is for organizations that need negotiated volume and long retention. A free trial is available, and paid plans begin at $99 per month.
Platforms
Traccia is offered through an API and web interface. Its SDK-based framework support works with Python and TypeScript agents, and hybrid deployment allows hosted services to be combined with customer-controlled components.
Who it's for
Choose Traccia when an agent platform team needs tracing, output evaluation and enforceable runtime limits in the same operating model. It is particularly useful for teams managing spend, approved models or recursion risk across multiple agent frameworks. It is a poorer fit for healthcare workloads that require a signed BAA, or for organizations seeking a completed SOC 2 Type II certification today.
Pros and cons
Pros
- Controls during execution: spend limits, model allowlists and loop caps can stop calls instead of merely reporting them.
- Agent-level visibility: sessions, tools, prompts and errors provide more useful context than request-only logs.
- Account-free export: the open-source SDK sends traces to established OpenTelemetry-compatible backends.
- Flexible growth path: five tiers, event-based quotas and declining overage rates accommodate projects from pilots to high volume.
Cons
- Short free retention: Hobby keeps traces for only seven days, limiting long-term comparisons.
- Compliance boundary: SOC 2 Type II is still in progress and no signed BAA is offered.
- Limited PII automation: masking is pattern-based; ML detection is not yet available.
- Provider-specific cost telemetry: automatic token and cost tracking is specified for OpenAI and Anthropic, not every LLM.
Alternatives
W&B Weave is a better choice if you want a free plan with AI-application evaluations, tracing and scorers plus broad desktop, web and self-hosted platform coverage.
SigNoz suits teams preferring self-managed infrastructure or a cloud observability product that also supports mobile clients.
Opik is worth choosing when running the core observability and evaluation stack as open-source code locally is the priority.
VoltOps offers an alternative freemium agent-observability option with a developer tier.
Helicone is an alternative for teams that specifically value an open-source platform and request-based usage allowances.
Maxim AI provides a free developer tier with up to three seats, one workspace, 10,000 logs per month and three-day retention.
PetalTrace is a free option for teams seeking modular agent-system tooling across web, desktop and self-hosted environments.
Respan is a fit when generous free quotas for logs, scores, datasets, evaluators and prompts matter more than Traccia's runtime policy controls.
For a wider shortlist, see AI Agent Observability Tools.
Verdict
Traccia is a strong choice for teams that need to see how agents run, evaluate outputs before release and enforce spend, model and loop policies at runtime. Its free plan lowers the barrier to a pilot, while higher tiers add the event capacity and retention that production operations require. Look elsewhere if a signed BAA is mandatory, if completed SOC 2 Type II certification is a requirement, or if you only need basic tracing without governance controls.
The verdict
Traccia combines agent tracing, evaluation, and policy controls, with a free plan and paid tiers from $99/month. Its HIPAA module does not make an organization HIPAA compliant, and Traccia says it does not currently offer a signed BAA.
Questions people ask
Is there a free Traccia plan?
Yes. Hobby is billed free and includes 50K events with seven days of retention.
How much do paid plans cost?
Paid plans start with Observe at $99 per month. Govern is $299 per month and Scale is $799 per month; Enterprise is custom priced.
Which integrations are listed?
The documentation lists LangChain, Claude Code, CrewAI, OpenAI Agents SDK, and Gemini SDK integrations.
Can Traccia trace any LLM?
Traccia says it works with any LLM. It automatically tracks tokens and costs for OpenAI and Anthropic, and uses OpenTelemetry tracing for HTTP-based providers.
Does Traccia provide a signed HIPAA agreement?
No signed Business Associate Agreement is currently offered. Enabling the HIPAA module alone does not make an organization HIPAA compliant.
Can traces be stored without a Traccia account?
Yes. The open-source SDK can export traces to Jaeger, Grafana Tempo, Zipkin, or any OpenTelemetry-compatible backend without an account.
What the maker publishes
- Data residency
- Production deployments can be configured for EU-region processing, with EU-only storage confirmed in the contract during onboarding. source
- Evaluations
- The platform supports prompts, datasets, scorers, and experiments for evaluating outputs before promotion. source
- HIPAA limitation
- Traccia says it does not currently offer a signed Business Associate Agreement and that enabling its HIPAA module does not make an organization HIPAA compliant. source
- Integrations
- The documentation lists integrations for LangChain, Claude Code, CrewAI, OpenAI Agents SDK, and Gemini SDK. source
- LLM coverage
- Traccia says it works with any LLM, with automatic token and cost tracking for OpenAI and Anthropic and OpenTelemetry tracing for HTTP-based providers. source
- Local SDK option
- The open-source SDK can export traces to Jaeger, Grafana Tempo, Zipkin, or any OpenTelemetry-compatible backend without an account. source
- Policy controls
- Spend limits, model allowlists, and loop caps can stop a call, while detective rules record after a run. source
- Privacy feature
- Optional pattern-based PII masking can run before spans leave the user's process; ML PII detection is planned. source
- Security status
- Traccia states that SOC 2 Type II certification is in progress and that granular role-based access control is available. source
- Support
- Product support for accounts, activation emails, integrations, and the hosted platform is available at [email protected]. source
- Tracing
- Its SDK captures agent runs, sessions, tool calls, prompts, and errors from Python and TypeScript. source
- What it does
- Traccia describes itself as an AI agent control plane for observability, evaluation, governance, and runtime policy enforcement. source
Checked against the maker's pages on 29 Sept 2026.