Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

W&B Weave

site.wandb.ai

  • Free plan
  • Open source
  • Windows
  • macOS
  • Linux
  • Web
  • iPhone and iPad
  • API
  • Self-hosted
8.8 Club score Visit site.wandb.ai

What W&B Weave is

W&B Weave provides observability and continuous improvement tools for production AI agents. It groups traces into sessions and turns, treating steps, tools, and sub-agents as distinct concepts. Built-in and custom signals capture and classify agent interactions, with alerts available through Slack notifications and webhook automations. Its evaluation framework compares results and visualizes potential regressions before deployment, while the Playground lets teams compare models and prompting approaches against production traces. Pre-built safety scorers cover toxicity, bias, personal information detection, and hallucinations; quality scores include coherence, fluency, and context relevance. Weave works with frameworks such as LangChain and LlamaIndex and models from OpenAI, Meta, Amazon Bedrock, and Anthropic. Python and TypeScript libraries are available, along with OpenTelemetry trace submission to its OTLP endpoint. The Free plan includes 1 GB per month of ingestion and 5 GB of storage. Pro costs $60/mo and includes 1.5 GB of monthly ingestion; additional Pro ingestion costs $0.10 per MB.

Who it suits

Weave suits teams developing or operating production AI agents who need trace monitoring, evaluation, and scoring. It also fits teams using its supported frameworks, models, or Python and TypeScript libraries.

Good to know, and what to watch

  • Groups agent traces into sessions and turns
  • Routes alerts through Slack or webhooks
  • Includes pre-built safety and quality scorers
  • Supports Python, TypeScript, and OpenTelemetry traces
  • Free plan includes 1 GB monthly ingestion
  • Pro includes 1.5 GB monthly ingestion
  • Additional Pro ingestion costs $0.10 per MB

Enterprise

—

Free

Free

Pro

$60/mo

Free plan · 30-day trial

Getting started

  1. Visit https://site.wandb.ai/weave/.
  2. Choose the Free plan, Pro, or an Enterprise plan.
  3. Use the 30-day trial if you want to evaluate a trial option.
  4. Connect through a Python or TypeScript library, or send OpenTelemetry traces to the OTLP endpoint.
  5. Use supported frameworks or models and review the listed integration options.

Limits to check

Free includes 1 GB per month of data ingestion and 5 GB per month of storage. Pro starts at $60 a month with 1.5 GB per month of ingestion; additional Pro ingestion costs $0.10 per MB.

The Club's review

W&B Weave is an observability and evaluation toolkit for production AI agents, aimed at teams that need to trace, assess and refine language-model applications. Its strongest case is the combination of agent-level tracing, evaluation and monitoring; ingestion allowances are the main cost to weigh against expected usage.

Overview

Weave is built to help teams follow AI application behavior and improve it continuously. It covers agent and retrieval tracing, LLM evaluations, prompt management, and cost and token tracking. Weights & Biases, founded in 2017 in San Francisco, was acquired by CoreWeave in May 2025.

Weave fits across AI Agent Observability Tools, AI LLM Observability Tools, AI Agent Evaluation Tools, Model Monitoring Software and LLM Observability Tools.

Key features

Agent tracing and monitoring

Weave groups traces into sessions and turns, with steps, tools and sub-agents treated as distinct concepts. That structure helps teams examine multi-stage agent activity rather than only individual model calls. Built-in and custom signals capture and classify interactions, and alerts can go to Slack or webhook automations, which gives teams options to connect monitoring to existing workflows.

Evaluation, experimentation and guardrails

The evaluation framework supports comparisons and visualizations to help identify regressions before deployment. In Playground, teams can compare models and prompting approaches against production traces. Pre-built safety scorers cover toxicity, bias, personal information detection and hallucinations; quality scorers assess coherence, fluency and context relevance. Together, these features connect evaluation to observed behavior, though the toolkit is most relevant to teams actively iterating on AI applications rather than those seeking only basic request logging.

Integrations, APIs and security

Weave works with frameworks including LangChain and LlamaIndex, and models from OpenAI, Meta, Amazon Bedrock and Anthropic. Python and TypeScript libraries are available, as is OpenTelemetry trace ingestion through an OTLP endpoint. W&B states that its platform holds ISO/IEC 27001:2022, ISO/IEC 27017:2015 and ISO/IEC 27018:2019 certifications and complies with SOC 2 Type 2 and HIPAA standards. Data is encrypted in transit with TLS 1.2+ and at rest with AES 256. Enterprise options include single sign-on, audit logs, secure private connectivity and customer-managed encryption keys.

Pricing

Weave uses a freemium model, with a 30-day trial and paid plans from $60/mo. The Free plan costs nothing and includes 1 GB/mo of Weave data ingestion, 5 GB/mo of storage, and AI application evaluations, tracing and scorers. It is a practical starting point for small workloads, but the monthly ingestion allowance can constrain teams collecting substantial production traces.

Pro starts at $60 a month, billed monthly, with annual billing also available. It is intended for organizations with fewer than 50 employees and raises ingestion to 1.5 GB/mo; additional ingestion costs $0.10/MB. Pro also includes priority email and chat support. That higher allowance may suit a growing team, but overages make usage forecasting important.

Enterprise has custom pricing and is invoiced annually upfront. It offers customizable usage, enterprise support, and security and compliance features, including controls such as single sign-on and customer-managed encryption keys. It is the relevant tier for organizations requiring those controls and support; the annual upfront term is a meaningful commitment. The Free and Pro ingestion caps are the main limitation for teams with high trace volumes.

Platforms

Weave is available through API, Linux, macOS, self-hosted, web and Windows environments. Both hosted and self-hosted deployment options are supported.

Who it's for

Weave is a strong fit for teams building production agents or other LLM applications that need a connected view of traces, evaluation results, prompts, retrieval and costs. Its evaluation comparisons, scorers and alerts are especially useful when teams need to catch behavioral regressions and route monitoring signals. Small projects can begin on Free; organizations with stronger security requirements can consider Enterprise.

It is less compelling for teams whose expected ingestion exceeds the plan allowances and that want predictable costs without per-megabyte Pro overages, or for those who need only a narrow logging tool rather than evaluation and improvement workflows.

Pros and cons

  • Pros: Sessions, turns, steps, tools and sub-agents are modeled explicitly, giving teams a structured way to inspect agent activity.
  • Pros: Evaluation comparisons, production-trace experimentation and pre-built safety and quality scorers support an improvement cycle before deployment.
  • Pros: Slack and webhook alerts, OpenTelemetry ingestion, and framework and model integrations connect Weave with common AI development workflows.
  • Cons: Free includes only 1 GB/mo of ingestion, while Pro includes 1.5 GB/mo and charges $0.10/MB for additional ingestion.
  • Cons: Enterprise is invoiced annually upfront, which may not suit organizations seeking a shorter commitment.

Alternatives

Choose Opik if running open-source observability and evaluation code locally is a priority. SigNoz offers a self-hosted community plan, with infrastructure, storage, scaling, upgrades and backups managed by the user. Choose Traccia if its event-based allowances and retention terms better match your needs: Hobby includes 50K events and 7 days of retention, while Observe costs $99 a month and includes 500K events.

VoltOps offers a Developer plan with one seat, one prompt, one project and 250 traces per month. Helicone is an open-source alternative with a Hobby plan that includes 10,000 free requests. Maxim AI offers a Developer plan with up to three seats, one workspace, up to 10k logs per month, three-day data retention and email support. PetalTrace is a free alternative. Respan (formerly Keywords AI) has a Free plan with 100k logs, 1k scores, five datasets, two evaluators and five prompts.

Verdict

Choose W&B Weave if your team needs agent-level observability tied directly to evaluation, scoring and alerts, especially when refining production AI behavior. Its breadth is the main reason to choose it; its modest monthly ingestion allowances and Pro overage charge are the main reasons to compare alternatives before committing.

The verdict

Weave combines agent tracing, evaluations, scoring, and alerts in one toolkit. Review the ingestion allowances and additional Pro ingestion charge against your expected data volume.

Questions people ask

Is W&B Weave free?

Yes. Free includes 1 GB per month of Weave data ingestion and 5 GB per month of storage.

How much does Pro cost?

Pro starts at $60 a month. Annual billing is also available, and additional ingestion costs $0.10 per MB.

Is there a free trial?

Yes, a 30-day trial is listed.

Which frameworks and models does Weave support?

It works with frameworks including LangChain and LlamaIndex and supports models from OpenAI, Meta, Amazon Bedrock, and Anthropic.

Can I send OpenTelemetry traces to Weave?

Yes. Weave supports sending OpenTelemetry traces to its OTLP endpoint.

What security standards does W&B report?

W&B states that its platform is certified to ISO/IEC 27001:2022, ISO/IEC 27017:2015, and ISO/IEC 27018:2019 and compliant with SOC 2 Type 2 and HIPAA standards.

What the maker publishes

Acquisition news
CoreWeave announced on 4 March 2025 that it would acquire Weights & Biases and completed the acquisition on 5 May 2025. source
Agent-native tracing
Traces are organized around sessions, turns, steps, tools and sub-agents as first-class objects. source
Apple app requirements
The W&B iPhone app requires iOS 18 or later, and its Mac version requires macOS 15 or later on Apple silicon. source
Behavior monitoring
Built-in and custom signals classify agent interactions, while Slack notifications and webhooks route detected issues. source
Company
Weights & Biases says it was founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis and was acquired by CoreWeave in May 2025. source
Company facts
LinkedIn lists Weights & Biases as privately held, founded in 2017, headquartered in San Francisco and employing 201–500 people. source
Deployment options
The deployment guide describes multi-tenant SaaS and single-tenant AWS Dedicated Cloud deployments for Weave. source
Documentation depth
The support center lists a dedicated Weave section with 16 articles and eight tags. source
Encryption
W&B states that data is encrypted in transit using TLS 1.2+ and at rest using AES 256. source
Enterprise controls
Enterprise pricing lists options including single sign-on, audit logs, secure private connectivity, and customer-managed encryption keys. source
Evaluations
Weave provides an evaluation framework with comparisons and visualizations to help identify regressions before deployment. source
Evaluations
An imperative evaluation API compares application or model changes with visualizations intended to expose regressions. source
Founded
2017 source
Guardrails
Pre-built safety scorers include toxicity, bias, PII detection, and hallucinations, while quality scores include coherence, fluency, and context relevance. source
Guardrails
Prebuilt safety scorers cover toxicity, bias, PII and hallucinations, while quality scorers cover coherence, fluency and context relevance. source
Headquarters
San Francisco, California, United States source
Integrations
W&B says Weave works with popular LLM frameworks including LangChain and LlamaIndex and supports models from OpenAI, Meta, Amazon Bedrock, and Anthropic. source
Integrations
The deployment guide lists integrations for OpenAI, Anthropic, Cohere, Mistral, LangChain, LlamaIndex, DSPy, Google Gemini, Amazon Bedrock, Together and Groq. source
Leaderboards
Evaluation results can be aggregated into leaderboards and shared across an organization. source
Legal name and ownership
Crunchbase identifies the legal entity as Weights & Biases, Inc., founded by Chris Van Pelt, Lukas Biewald and Shawn Lewis, and records seven funding rounds and acquisition by CoreWeave. source
Licence
The public Weave repository is released under the Apache-2.0 license. source
Limit: Apple app access
An App Store reviewer complained that meaningful functionality is gated behind a paid multi-tenant account. source
Limit: G2 complexity and cost
G2 reviewers repeatedly mention an overwhelming interface, documentation or guidance gaps, sluggish charts on large projects and rapidly rising storage or per-user costs. source
Limit: independent review pricing
Aicoolies says ingestion is usage-based and can make total cost difficult to estimate before a representative trial. source

Checked against the maker's pages on 20 Oct 2026.