Orange Collective
Mastra

Mastra

The open-source TypeScript framework for building AI agents, from the team behind Gatsby.

mastra.aiYYC ProfileGitHub@mastra_aiLinkedIn
Mastra launch demo · YC W25[10]

25k

GitHub stars

Passed vercel/ai in June 2026 · Apache-2.0

4M/mo

npm downloads

@mastra/core, May 2026 — 9k in Jan 2025

$35M

Total raised

$22M Series A led by Spark Capital, Apr 2026

In production at

Replit

AI coding agent

Salesforce

Enterprise reference · Series A post

Workday

Enterprise reference · Series A post

PayPal

In production at 1.0 launch

MongoDB

Enterprise reference

Brex

Fintech agents

Marsh McLennan

Enterprise search · 100k+ daily users

Enterprise customers named in the Series A announcement, plus broad adoption across the W25, S25, and F25 YC AI cohort — Mastra is the JS-first default for agents.[2] [18] [20]

Thesis

Python is where models are trained; TypeScript is where most product code that calls them gets written. Most developers shipping AI features in 2026 are not opening a notebook — they're already inside a Next.js project and reaching for the framework that feels native to it.[6] [7] Mastra is that framework, written by the team that built Gatsby.[2] [4] The longer-term bet: the same OSS distribution motion that put React-era frameworks in front of millions of web teams now runs on agent orchestration — by the operators who ran it once already. Sixteen months on, the bet is compounding: 25k GitHub stars, ~4M monthly npm downloads, a 1.0 release, and a $22M Series A led by Spark Capital.[16] [17] [18]
  1. 01

    Most AI app developers already write TypeScript. 72% of full-stack developers already use TypeScript; only 29% of web teams claim Python proficiency.[5] Point72 Ventures puts it bluntly: TypeScript is "becoming the language of choice for AI application development," and web developers — not ML engineers — will lead AI app implementation at scale.[6] Mastra is built for the audience that's already there, not the one a Python framework hopes to convert.

  2. 02

    The frontier moved from rendering pages to orchestrating agents. React, Next.js, and Gatsby fought over rendering and routing. Agent code needs different primitives — steps, tools, memory, RAG, evals — and Mastra ships those as TypeScript-native APIs — agents, workflows, memory, rag, evals — not a Python port stitched into the JS world.[3] [11]

  3. 03

    The founders shipped Gatsby. They know how to win OSS distribution. Sam Bhagwat (CEO, co-founder of Gatsby), Abhi Aiyer (CTO, ex-Netlify principal eng), and Shane Thomas (CPO, ex-Gatsby head of product) scaled Gatsby to 55k+ stars, 4k+ contributors, and $5M ARR before the Netlify acquisition.[2] [4] The same OSS distribution playbook, now running on a much larger category — every web team building any AI feature, vs. the static-site subculture that bought Gatsby.

  4. 04

    Coding agents are the new distribution layer. ChatGPT, Cursor, and Claude Code recommend whatever has the cleanest OSS repo, the most idiomatic SDK, and the simplest install. Apache-2.0, npm create mastra@latest, a star curve that went 1.5k → 10k in the four weeks around the YC launch and 25k by June 2026, and TypeScript-native by default is exactly the shape coding agents reach for.[13] [16]

Problem

Building a production agent in JavaScript today means stitching together five OSS libraries written for someone else.

Vercel's AI SDK is a thin streaming wrapper. LangChain.js is a port of a Python framework with the wrong idioms — chains instead of typed graphs, callbacks instead of suspend/resume. Memory, RAG, evals, and orchestration each live in their own repo, each with its own opinions, each with its own bug list.

Sam Bhagwat's framing of the gap is direct: "We noticed our friends building AI applications getting stuck debugging prompts, figuring out why their agents called (or didn't call) tools, and writing lots of custom memory retrieval logic."[10] Every senior JS engineer at every YC AI company today wants to ship an agent. The friction is real — and the only existing frameworks are either Python-shaped or too thin to matter.

The cost isn't the stitching. It's the chilling effect on what gets shipped. The team that has to integrate four libraries and write its own memory layer ships an agent in a quarter. The team that runs npm create mastra@latest ships one in an afternoon.

72%

of full-stack devs use TypeScript

vs. 29% who claim Python proficiency in web teams

98%

YoY growth in genAI projects

GitHub Octoverse 2024 · application layer is exploding

15%

YoY npm consumption growth

JavaScript ecosystem still expanding faster than any other

PureCode TS/Python comparison[5] · GitHub Octoverse 2024[7]

Why Now

Three preconditions converged in the same eighteen months.

TypeScript took the application layer, AI moved from research into product, and coding agents started picking the stack. The JS-native agent framework wasn't possible — or necessary — before now.

We noticed our friends building AI applications getting stuck debugging prompts, figuring out why their agents called (or didn't call) tools, and writing lots of custom memory retrieval logic.

Sam Bhagwat

Sam Bhagwat[10]

Co-founder & CEO · Mastra

There are at least as many JavaScript developers as Python developers, so now tools are increasingly catering to this widely expanded audience.

Shawn Wang

Shawn Wang[6]

swyx · Latent Space

TypeScript is becoming the language of choice for AI application development. Web developers — not ML engineers — will increasingly lead AI application implementation at scale.

P72

Point72 Ventures[6]

Perspectives · AI Application Layer

Each precondition on its own is interesting. Together they're a new category.

TypeScript took the application layer. TypeScript adoption tripled from 12% (2017) to 34% (2023) and overtook Java to enter GitHub's top three languages. npm consumption grew 15% YoY in 2024 — the JS ecosystem is still expanding faster than any other.[6] [7] Every modern dev tool ships TS types first.

AI moved from research into product. Generative AI projects grew 98% YoY on GitHub.[7] The dominant model is no longer "a researcher writes a notebook" — it's "a Next.js engineer ships an agent feature inside a SaaS app." The audience and the tooling have inverted.

Coding agents pick the stack. Developers ask ChatGPT how to ship an agent. ChatGPT, Cursor, and Claude Code recommend whatever has the cleanest OSS repo, the most idiomatic SDK, and the most-cited examples. Apache-2.0, a one-line install, and 25k stars is exactly the surface area coding agents are tuned to surface — and the download curve confirms the loop is running: @mastra/core grew from 9k monthly downloads in January 2025 to ~4M in May 2026.[13] [16] [17]

Web developers — not ML engineers — will increasingly lead AI application implementation at scale. TypeScript is the natural language for that work.
— Point72 Ventures[6]

How It Works

One framework. Six primitives. The TypeScript stack a JS engineer reaches for to ship an agent in an afternoon.

Step 01

Agents + model routing

Autonomous loops with tool calling, persistent state, and built-in routing across 90+ model providers via the Vercel AI SDK. One typed configuration object replaces a half-day of LangChain.js boilerplate.

Step 02

Workflows + human-in-the-loop

Graph-based control flow with .then(), .branch(), .parallel(). First-class suspend/resume for human-in-the-loop steps and state persistence. Mirrors XState patterns the JS world already knows.

Step 03

Memory + RAG + evals

Working memory, semantic recall, vector retrieval, and built-in evaluation harnesses as first-class primitives. No four-library handshake to keep state — and no separate eval framework to bolt on at the end.

The local dev playground is where the addiction starts.

Install in one command. npm create mastra@latest scaffolds the project, drops a local dev server with traces and logs, and gives the developer an agent calling a model inside ninety seconds. That's the Vercel-era expectation — and the bar Python frameworks structurally can't meet.[13]

Production observability, day one. OpenTelemetry tracing, eval metrics, and a real-time dashboard shipped in the same package. Compatible with Datadog, New Relic, and the broader OTel ecosystem the rest of the JS world already uses.[11]

Deploy anywhere a Node runtime runs. Vercel, Cloudflare Workers, Netlify, AWS, self-hosted — whatever the team's existing JS deploy target is. The agent ships alongside the rest of the app, not as a separate Python microservice the JS engineer can't debug.

The Framework Wars Are Now About Agents

React won components. Next.js won rendering. Mastra is the bet on what comes next.

The last generation of frameworks fought over rendering and routing. The next one is fighting over agent orchestration, and the OSS distribution playbook that worked last time still fits the shape of the problem.

Different primitives. The same OSS distribution motion that worked last time.

React shipped components. Next.js shipped routes and SSR. Gatsby shipped a build pipeline. Mastra ships agent primitives — steps, tools, memory, RAG, evals. The category is different; the framework shape is recognizable: a TypeScript-first DX, opinionated primitives, and a sharp opinion about what belongs in the framework vs. outside it.[3]

The distribution motion is the part the team has run before. OSS-first, permissive license, GitHub stars as the leading indicator, paid cloud runtime as the monetization vector — Gatsby ran this play to 55k stars and $5M ARR before the Netlify acquisition.[4] Vercel ran a related play with Next.js to a $3.25B valuation.[8] LangChain ran it in Python to a $1.25B Series B in October 2025 — on a reported $12–16M of ARR, which is what the market currently pays for the category leader's monetization curve.[21] Mastra is running it again, into a category — every web team building any AI feature — that is materially bigger than the static site subculture that bought Gatsby.

The hard, non-obvious work is the part the Gatsby team has already done once: community management, contributor scaling, ecosystem building, and the painful inflection between "great OSS project" and "profitable framework company." That's the reason to underwrite the bet — not any single primitive in the SDK.

0 → 25k GitHub stars in 22 months — and past Vercel's AI SDK repo

Chart

Sampled stargazer timestamps. The repo opened Aug 2024; the YC W25 launch took it from 1.5k to 10k inside roughly four weeks (Feb–Mar 2025). Steady compounding since: 20k at the 1.0 release (Jan 2026), 24,989 on Jun 12, 2026 — edging past vercel/ai (24,824).

Source · GitHub API, stargazer sampling · Jun 12, 2026

GitHub API stargazer sampling, Jun 12 2026[16] · vercel/ai comparison[22]

The Gatsby team didn't get to 55k stars by accident, and they're not running this playbook again casually. When operators with a proven OSS distribution motion pick a category that's materially bigger than their last one, the prior is they reach a serious outcome.
— Orange Collective

Market

The densest buyer pool is every YC AI batch — and beyond that, every web team.

Inside YC, Mastra is already the JS-first default — adoption spans W25, S25, F25, and P26 batches, with companies like Autumn (S25), Fireworks AI, and dozens of others running Mastra-powered agents in production.[1] [15] The dense YC AI cohort is exactly the acquisition channel that compounds — every YC company in production is a forcing function for the next batch's choice. And the buyer pool stopped being startups-only in 2026: the Series A announcement named Salesforce, Workday, MongoDB, Brex, Indeed, and Marsh McLennan — whose Mastra-built internal search serves 100k+ employees a day — as customers.[18]

Beyond YC, the addressable market is every web team building AI features. The TypeScript developer base is 17M+ globally and growing 15%+ YoY.[6] [7] Every web team eventually ships an agent feature; very few will context-switch their stack to Python to do it. The framework that wins JS wins the application layer.

Near term — JS-first AI startups and YC AI cohorts

YC current and recent AI cohorts plus the broader JS-native seed-to-Series-A pool. Dense network, technical buyers, OSS-friendly. Mastra is already the JS-first default for agents — and the YC cohort dynamic is the strongest distribution channel in software.[2] [15]

Long term — every web team building AI features

The 17M+ TypeScript developer base growing 15%+ YoY is the long pole.[7] Every web team eventually ships an agent feature. The AI infrastructure category is on a path to a $200B+ market by 2030 — and the framework layer is the one with developer leverage and ecosystem lock-in.

Every YC AI company is an agent framework decision in waiting. Mastra should be the answer by default — and that's how the next decade of AI application infrastructure gets written.
— Orange Collective

Traction — June 2026 update

Sixteen months after this memo: a 1.0 release, two rounds, and a ~440x download curve.

We wrote this memo in February 2025 at ~1.5k GitHub stars. The milestones since are the reason we keep underwriting the position.

OCT '25

$13M seed — YC-led, 120+ investors

Paul Graham, Guillermo Rauch (Vercel), Amjad Masad (Replit), Arash Ferdowsi (Dropbox), Gradient Ventures, SV Angel, Runa Capital — and Orange Collective. The investor list doubles as a distribution map of the JS ecosystem.[19]

JAN '26

Mastra 1.0 stable

v1 beta shipped Nov 6, 2025; 1.0 stable landed Jan 21, 2026 at 300k+ weekly npm downloads, with Replit, PayPal, Sanity, Brex, and SoftBank in production. The framework froze its public API the way Next.js did before its growth phase.[20]

APR '26

$22M Series A + the paid platform

Led by Spark Capital, $35M total raised. Shipped alongside it: hosted Mastra Studio (evals, traces, datasets), Mastra Server, and Memory Gateway — the monetization layer the Gatsby playbook always pointed at. Named customers: Salesforce, Workday, MongoDB, Brex, Indeed, Marsh McLennan.[18]

@mastra/core npm downloads: 9k → 4M per month in 17 months

Chart

Monthly registry downloads, January 2025 through May 2026 (last full month). The curve is still convex — each of the last four months added more absolute volume than the entire first year. Weekly run-rate stood at ~887k in mid-June 2026.

Source · npm registry API · api.npmjs.org · Jun 12, 2026

npm registry range API[17] · weekly point-in-time[22]

Mastra announcement card: We raised a $22M Series A to help every developer build agents — Sam Bhagwat, Apr 9, 2026
Series A announcement · Spark Capital · Apr 9, 2026[18]

Competitive landscape

Four categories of competition. Mastra is positioned against all of them.

Each category has a structural limitation — language idiom, scope depth, or vendor lock-in. Mastra's OSS + TypeScript-native + full-stack stance is the answer to all four.

LangChain / LangGraph

Python incumbent · $1.25B

139k stars, $125M Series B at $1.25B (IVP, Oct 2025) on a reported $12–16M ARR.[21] The center of gravity moved to LangGraph — 2.56M weekly npm downloads for the JS package, still ahead of Mastra's 0.89M.[22] But the idioms remain ports from Python, and Mastra's curve is the one compounding: ~440x monthly download growth since Jan 2025.[17]

Vercel AI SDK

The layer below — mostly

24.8k stars, 12.45M weekly downloads — the de facto model interface for JS.[22] AI SDK 6 added a lightweight Agent abstraction, tool approval, and MCP support[24] — a real scope expansion, but still no durable workflows, memory service, RAG, or evals. Mastra builds on the AI SDK and sells the stack above it; Rauch backed Mastra's seed.[19]

OpenAI Agents SDK (TS)

Vendor toolkit · Jun 2025

OpenAI shipped a TypeScript Agents SDK in June 2025 — now at ~1M weekly downloads.[23] [22] It validates the TS-agents category and is lock-in by design. Mastra routes across 90+ providers through one interface, so the same agent code runs on whatever model wins the quarter.[11]

LlamaIndex / CrewAI / Pydantic AI

Python research stack

Strong for retrieval and multi-agent research (CrewAI at 53k stars, Pydantic AI at 18k).[16] Wrong idiom for shipping a customer-facing product on a JS stack — LlamaIndex's TS package does ~120k weekly downloads, a rounding error next to the JS-native options.[22]

Weekly npm downloads, agent-framework packages — June 2026

Chart

Vercel's ai package is the model-interface layer most frameworks (including Mastra) sit on, so its 12.4M is the ecosystem ceiling, not a direct competitor count. Among full agent frameworks, Mastra at 0.89M weekly is within 3x of LangChain's JS packages despite launching two years later — and growing faster than every bar to its left.

Source · npm registry API, week ending Jun 11, 2026

npm registry point-in-time API[22]

LangChain's gravity is Python's gravity. Vercel's scope is the UI layer. OpenAI's framework is model-vendor lock-in. The opening in the middle — TypeScript-native, full-stack, model-agnostic — is exactly where Mastra is building.
— Orange Collective

Founder deep dive

Three founders, one prior framework with 55k stars, and a chip on the shoulder about what could have gone bigger.

Why Sam built it. Sam co-founded Gatsby in 2017 and spent eight years scaling it to 55k+ stars, 4k+ contributors, and $5M ARR before the Netlify acquisition.[2] [4] He learned every lesson the hard way: how to win a JS framework war on the merits, how to monetize an OSS community without alienating it, and — by his own framing — what the team didn't do well enough the first time. Self-described as "hungry with a chip on our shoulder." Mastra is the next at-bat on a 10× bigger market.[10]

Why Abhi built it. Abhi was principal engineer at Netlify post-acquisition, leading a 100+ person engineering org and running the Gatsby Cloud infrastructure that served billions of files and hundreds of TB of bandwidth a month. He knows what it takes to operate a developer framework at the scale where every regression is a public incident — and what kind of architecture survives that pressure.

Why Shane built it. Shane was staff engineer and head of product at Gatsby, with 15+ years in open source. He's also the founder who shipped Audiofeed, an AI podcasting tool — so he was inside the agent-building problem before the team formalized Mastra. State wrestling champion, which is unrelated to the framework but is the kind of detail that tells you what kind of operator he is.

Why this team is the right team. An OSS-framework CEO who has done the full GTM cycle from zero to acquisition. A CTO who has scaled framework infrastructure to billions of requests. A CPO who has shipped both the OSS contributor experience and an AI product. They cover every leg of the framework-company stool: distribution, infra, and DX. They've built one framework company together before; the second one is being built faster, on a bigger market, with the operating mistakes already paid for.

Why velocity is a feature. The repo opened in August 2024, went from 1.5k to 10k stars in roughly the four weeks around the YC W25 launch, and passed Vercel's own AI SDK repo at ~25k in June 2026.[16] The team shipped v1 beta in November 2025 and froze the public API at 1.0 stable in January 2026 — fourteen months of monthly primitive releases without losing the cleanest docs surface in the category.[20] That velocity is the lead indicator of whether the OSS playbook compounds.[14]

The long arc. Mastra becomes the framework every JS engineer reaches for the moment "agent" enters the PRD. The OSS core wins the install. The cloud runtime powers the renewal. The enterprise observability and control plane absorb the next layer of spend. Every JS team building any AI feature ends up inside one platform — and the operational memory of how thousands of agents are built and shipped is the long-term moat.

Founder & team

Sam Bhagwat

Sam Bhagwat

Repeat FounderExited

Co-founder & CEO

Co-founder of Gatsby.js — open-source React framework, scaled to 55k+ stars and $5M ARR before the Netlify acquisition in 2023. Early engineer (<20) at Zenefits (W13) and PlanGrid (W12). Stanford '11. Spent two years door-knocking for political campaigns before tech. Self-described as "hungry with a chip on our shoulder" — Gatsby taught the team what works and what doesn't in OSS distribution, and Mastra is the next at-bat.

Abhi Aiyer

Abhi Aiyer

Repeat FounderExited

Co-founder & CTO

Principal engineer and lead of a 100+ person engineering org at Netlify, post-Gatsby acquisition. Built Gatsby Cloud infrastructure — tens of thousands of build nodes, billions of files served, hundreds of TB of monthly bandwidth. Knows what it takes to operate a developer framework at the scale where every regression is a public incident.

Shane Thomas

Shane Thomas

Repeat Founder

Co-founder & CPO

Staff engineer and head of product at Gatsby. 15+ years in open source. Previously built Audiofeed (AI podcasting). State wrestling champion — turns out, useful preparation for shipping an OSS framework into a category dominated by a Python incumbent with 100k stars.

Risks & mitigations

Risk

LangChain has a multi-year head start, 139k+ stars, a $1.25B Series B (IVP, Oct 2025), and is the answer when an AI engineer Googles "agent framework." On npm itself, @langchain/langgraph still out-downloads @mastra/core roughly 3:1 (2.6M vs. 0.9M weekly, June 2026).

Mitigation

LangChain's gravity is real, but its center of mass is Python — and its $1.25B valuation on a reported $12–16M ARR is also the comp that prices what this category is worth. LangChain.js carries ported idioms into JS (chains over workflows, callbacks over typed graphs), and a large share of its npm volume is legacy transitive dependencies from the 2023–24 LLM-app wave. Mastra is the only TypeScript-native option built around patterns JS engineers already know, and the growth differential is the tell: @mastra/core went from 9k to ~4M monthly downloads in seventeen months while crossing 25k stars — past Vercel's own AI SDK repo. The gap is closing from the side that's compounding.

Risk

Vercel is the closest neighbor — and AI SDK 6 shipped an Agent abstraction, tool-approval flow, and MCP support. The "Vercel won't ship agents" assumption from our original memo has partially expired.

Mitigation

Partially materialized, and worth saying plainly. AI SDK 6's Agent is a lightweight loop — define a model, instructions, tools — not a platform: no durable workflow engine, no memory gateway, no RAG pipeline, no evals/observability product. Mastra still routes models through the AI SDK and sells everything above it, which is where Mastra Studio, Server, and Memory Gateway (launched with the Series A, Apr 2026) now sit. The relationship is also not arms-length: Guillermo Rauch personally backed Mastra's seed. The risk to keep pricing is scope creep in AI SDK 7+, not the current release.

Risk

OSS monetization risk — Gatsby itself never broke $10M ARR before the Netlify acquisition. Will the same playbook scale this time?

Mitigation

The monetization surface is no longer hypothetical: Mastra shipped its paid platform — hosted Studio (evals, traces, datasets), Mastra Server, and Memory Gateway — alongside the $22M Series A (Spark Capital, Apr 2026), with Salesforce, Workday, MongoDB, Brex, Indeed, and Marsh McLennan named as customers. Marsh McLennan's Mastra-built internal search serves 100k+ employees daily. And the LangChain comp says the market pays up for exactly this motion: $1.25B on $12–16M ARR. The runtime is usage-based in a way Gatsby's build platform never was.

Risk

The AI framework category is moving fast — primitives that look canonical today (agents, workflows, RAG) could be obsoleted by the next model generation. OpenAI now ships its own TypeScript Agents SDK.

Mitigation

Model providers keep absorbing capabilities — OpenAI's TS Agents SDK (June 2025) reached ~1M weekly downloads in a year, which validates the category while staying lock-in by design. Agent orchestration, memory, observability, and human-in-the-loop are the layers that stay above the model line and stay model-agnostic. Mastra sits exactly there, shipped 1.0 stable in January 2026 after fourteen months of monthly primitive releases, and routes 90+ providers — so the same agent code survives whichever lab wins the quarter. The defensible asset is the TypeScript-first audience and the Gatsby team's OSS distribution muscle, not any single primitive in the SDK.

What we're watching

  • Whether @mastra/core's download curve closes the gap with @langchain/langgraph — 0.9M vs. 2.6M weekly today, but Mastra's monthly volume grew ~440x since Jan 2025 while crossing 25k stars and passing vercel/ai. A crossover inside 12 months is the strongest possible signal.
  • Series A platform conversion — Mastra Studio, Server, and Memory Gateway (Apr 2026) turning the Salesforce / Workday / PayPal / Marsh McLennan logos into reportable ARR. LangChain printed a $1.25B valuation on $12–16M ARR; Mastra's revenue line is the next thing to underwrite.
  • Vercel AI SDK 7 scope — AI SDK 6 shipped a lightweight Agent abstraction. If the next major release adds durable workflows or memory, the partnership-shaped boundary between the two projects gets renegotiated.
  • Coding-agent recommendation share — when Cursor and Claude Code generate agent code, what fraction picks Mastra vs. LangChain.js, OpenAI's TS Agents SDK, or raw Vercel AI SDK.

References

  1. [1]Mastra — Product homepage
  2. [2]Y Combinator — Mastra company profile (W25, 30 employees, SF)
  3. [3]GitHub — mastra-ai/mastra (Apache-2.0, 25k+ stars, 2.2k+ forks)
  4. [4]GitHub — gatsbyjs/gatsby (55k+ stars, 4k+ contributors — prior OSS framework by the same team)
  5. [5]PureCode — TypeScript vs. Python developer demographics
  6. [6]Point72 Ventures — TypeScript: enterprise-ready and the ideal option for AI application development
  7. [7]GitHub Octoverse 2024 — TypeScript adoption, npm consumption +15% YoY, generative AI projects +98% YoY
  8. [8]TechCrunch / public filings — LangChain $25M Series A (Sequoia, Feb 2024); Vercel $250M Series E (Accel, May 2024)
  9. [9]Netlify — State of Web Development / framework popularity
  10. [10]Mastra Launch — YC Launch post ("If you hate Langchain, you'll love Mastra")
  11. [11]Mastra docs — Agents, workflows, RAG, memory, evals, model routing (90+ providers)
  12. [12]Netlify — State of the JAMstack 2020 (Gatsby usage at enterprises)
  13. [13]Mastra — npm create mastra@latest install (single-command setup)
  14. [14]Mastra Discord — community channel referenced in repo README
  15. [15]Orange Collective — internal portfolio (Mastra is in production at multiple OC portfolio companies, including Autumn; OC participated in the Oct 2025 seed)
  16. [16]GitHub API — mastra-ai/mastra stargazer history (24,989 stars as of Jun 12, 2026; trajectory sampled from starred_at timestamps)
  17. [17]npm registry API — @mastra/core monthly downloads, Jan 2025 – Jun 2026 (9,018 → 3,966,414/month)
  18. [18]Mastra Blog — "We raised a $22M Series A" (Spark Capital, Apr 9, 2026; $35M total; Mastra Studio, Server, Memory Gateway; Brex, Sanity, Indeed, Marsh McLennan, MongoDB, Workday, Salesforce, Replit customers)
  19. [19]TechNews180 — Mastra raises $13M seed (Oct 2025; YC-led, 120+ investors incl. Paul Graham, Guillermo Rauch, Amjad Masad, Gradient Ventures, SV Angel, Orange Collective)
  20. [20]Mastra Blog — Announcing Mastra v1 beta (Nov 6, 2025); 1.0 stable shipped Jan 21, 2026 with 300k+ weekly npm downloads
  21. [21]TechCrunch — LangChain hits $1.25B valuation with $125M Series B led by IVP (Oct 2025; reported ARR $12–16M)
  22. [22]npm registry API — weekly downloads, week ending Jun 11, 2026: ai 12.45M · @langchain/langgraph 2.56M · langchain 2.41M · @openai/agents 1.03M · @mastra/core 0.89M · llamaindex 0.12M
  23. [23]GitHub — openai/openai-agents-js (OpenAI Agents SDK for TypeScript, launched Jun 2025)
  24. [24]Vercel — AI SDK 6 announcement (Agent abstraction, tool-execution approval, DevTools, MCP support)