Mastra 推出定制化套件计划
Introducing the Mastra Custom Harness Program
Mastra 发布定制化套件计划，工程师将嵌入客户团队共同构建满足其目标的套件；Workday 迁移至 Mastra 仅用 31 天，减少 3,000 行代码并将向量存储成本降低 98%；客户可避免供应商锁定，获得可衡量的投资回报。
Own your agent system, unlock from vendors, and control how work gets done. Our forward-deployed engineers co-build your harness on Mastra, tailored to your goals.
This year, as models got better, harness demand grew. Agents delivered value and sophisticated ones invested in harnesses for even better results.
In August, Salesforce shipped a harness for all non-Anthropic models to its 100K+ dev community, built on Mastra.
MongoDB built an internal platform, Sage, so every engineering team could ship Mastra agents. Workday's security team needed to triage alerts to their NOC and built a custom harness SRE with Mastra.
So today we're launching the Mastra Custom Harness Program. Our forward-deployed engineers embed with your team and co-build your harness, tailored to your goals.
Why teams build their own
You own the system
MongoDB's first agent was hand-rolled Go. Then they rebuilt it on Mastra and made the plumbing their own in Sage. Now release notes, CI triage, and support routing run on it. A well-designed harness becomes a platform for the whole company.
Avoid vendor lock-in
Workday spent months in production on Amazon Bedrock Agents. Moving to Mastra took 31 days, cut 3,000 lines of code, and lowered vector-storage costs by 98%. Mastra is open source, so you can swap models, compose your stack, and inherit every improvement we ship.
"After playing around with it for like a week, we were like: Mastra can do everything we need." — Bryce Montano, Engineer, Workday Security
Take control when you need
A custom harness gives you control. Marsh McLennan built LenAI, an agent platform on its own infrastructure, to keep company data out of consumer chat tools. Salesforce's harness pauses for human approval before sensitive tool calls.
Measurable ROI at scale
Workday's bar: triage alerts before human analysts. The agent does it in sixty seconds, which saves a person-hour every three alerts. MongoDB ties the cost of every agent request to the time it saves a DevOps engineer. A few dollars spent on an agent analyzing logs saves an hour of reading.
What you get with Custom Harness
- Dedicated FDEs in your repo and your Slack, bringing patterns from teams that have already shipped this.
- Observability, evals, and scoring, wired into your workflows from day one.
- Custom MCP, durable human-in-the-loop workflows, and tools for teams who will never open an IDE.
- Auth, RBAC, and governance controls, ready for your security review.
- Early access to the Mastra Enterprise Platform. See what we ship before it ships.
- An outcomes-focused engagement. We'll ship an agent system into production your users will adopt that's designed to deliver on your business goals.
Let's whiteboard it
Bring your goals, workflows, stack, and constraints. Our FDEs will sketch the architecture with you by hand, showing you what we'll build, what we'll leave alone, and what we'll pull in from Mastra.
Schedule a live whiteboarding session with our FDEs.
Sam Bhagwat is the founder and CEO of Mastra. He co-founded Gatsby, which was used by hundreds of thousands of developers. A Stanford graduate and veteran of web development, he authored 'Principles of Building AI Agents' (2025).
AI agent architecture: patterns, components, and production design
ArticleAug 9, 2026 Learn what an AI agent builder is, the core building blocks of an AI agent, and how to build, test, and deploy your first agent to production. Learn what an AI hallucination is, why language models produce false outputs, and practical techniques to detect and reduce hallucinations.Agent builder: how to build, test, and deploy AI agents
AI hallucination: what it is, why it happens, and how to reduce it
来源: Mastra Blog · mastra.ai