Vellum pricing

About

Vellum started life as something else entirely. Founded by Akash Sharma, Noa Flaherty, and Sidd Seethepalli (who built LLM features together at Dover, YC S19), it launched out of Y Combinator’s W23 batch as a developer platform for building production LLM applications — prompt engineering, evaluations, deployments, monitoring. It raised a $5M seed in July 2023 (Rebel Fund, Eastlink Capital, Pioneer Fund, Y Combinator) with 40 paying customers, and a $20M Series A led by Leaders Fund in July 2025, by which point it counted Swisscom, Redfin, Drata, and Headspace among customers of its enterprise AI workflow platform.

Then, on May 7, 2026, Vellum relaunched as a personal AI assistant — “Your own Personal Intelligence” — announcing $25M raised from Dharmesh Shah, Arash Ferdowsi, Rebel Fund, and Y Combinator. The new product is an MIT-licensed open-source assistant (25K+ commits on GitHub) with persistent associative memory, its own identity, and proactive reach-outs across macOS, Telegram, and Slack. The hosted version is priced with a free Base plan, a Pro tier available either as three preset packages (Mighty, Super, Ultra) or a self-assembled Custom build (platform fee plus machine and storage tiers), and pay-as-you-go credits for AI usage beyond any included allowance. No public announcement documents what happened to the enterprise platform; its legacy product pages remain live, but the site is now entirely repositioned around the assistant.

For the most current information, visit Vellum.

Pricing summary : How Vellum’s pricing model works

Vellum prices the hosted assistant like a tiny cloud provider, not a chatbot subscription. The Base plan is free — no credit card — and includes a Small machine (1 vCPU, 2 GiB RAM), 6 GiB of persistent storage, and managed LLM credentials. Pro now splits into two paths: three preset packages — Mighty $30/mo (Small machine, 10 GiB storage, $25/mo credits included, no platform fee), Super $100/mo (Medium machine, 30 GiB, $45/mo credits included, platform fee + subdomain included), and Ultra $200/mo (Large machine, 60 GiB, $115/mo credits included, platform fee + subdomain included) — or a Custom build: a $10/mo platform fee plus a machine tier (Medium +$35, Large +$60, XL +$125/mo) plus a storage tier from 10 GiB (+$5/mo) to 120 GiB (+$30/mo — the previous 250/500 GiB tiers are now grandfathered to existing subscribers only) and an optional recurring usage bundle — Mighty (+$25/mo), Super (+$45/mo), or Ultra (+$115/mo), the same three amounts bundled into the preset packages. The minimum Custom build is $50/mo (fee + Medium + 10 GiB, no bundle); editing any single tier on a preset package automatically converts it to Custom.

AI consumption rides on a second meter: credits, where $1 = 1 credit, consumed by LLM inference, web search, image generation, and paid third-party APIs. Preset packages each include a monthly usage allowance (named for the package); beyond that — and on Base, or a Custom build without a bundle — credits are pay-as-you-go, and Vellum states it passes LLM costs through at cost, with no markup. Included usage resets each billing cycle with no rollover, while purchased pay-as-you-go credits expire 12 months after purchase.

As of August 2026, the Custom plan’s optional add-on was renamed from “credit bundle” to “usage bundle” and its menu narrowed from five flexible price points ($10/$25/$50/$100/$200/mo) to exactly the three preset-matched amounts (Mighty $25, Super $45, Ultra $115/mo); subscribers who already had a bundle at another amount keep it.

What makes this different: Vellum deliberately takes zero margin on inference and monetizes compute, storage, and convenience instead — and because the assistant is MIT-licensed and free to self-host, the hosted plans sell hosting rather than the capability itself.

Pricing by product

Sales motions across products: entirely self-serve PLG — free Base signup, self-serve Pro upgrades and credit top-ups, and an open-source self-host path. The current pricing page has no sales-led tier at all, a sharp break from the 2024-2025 quote-only era.

Hidden costs : What Vellum users actually pay

The subscription is the predictable half of a Vellum bill; credits are the variable half, and they accrue even when you are not chatting. Background operations — memory consolidation, conversation summarization, hourly heartbeat check-ins, and filing agents — all consume credits, though Vellum lets you reduce their frequency or route them to cheaper models. Credits are prepaid in $10-$100 increments, with optional auto-reload below a threshold you set ($1-$100) and a monthly spending cap ($25-$10,000) as a guardrail.

Want to estimate your own Vellum bill? Use the Vellum pricing calculator to model your costs based on usage patterns.

Pricing evolution : Vellum pricing history and changes

Cadence

Tracked range: 2023-present, via Wayback Machine snapshots (2023-06, 2024-08, 2025-07, 2025-12, 2026-03) and live captures on 2026-06-10, 2026-07-29, and 2026-08-28.

Notable changes

• 2023-06 — Launch-era public pricing for the LLM dev platform: Free $0 / Starter $49 per user/mo / Growth $249/mo / Pro $699/mo / Enterprise custom, limited by LLM production requests per week, deployments, and seats.
• 2024-08 — Every dollar figure removed. The page becomes quote-only (Growth / Pro / Enterprise, all ‘Request Demo’) as Vellum chases enterprise AI teams.
• 2025-07-10 — $20M Series A led by Leaders Fund (with YC, Socii Capital, Rebel Fund, Pioneer Fund, Eastlink Capital). Pricing stays gated.
• 2025 H2 — Self-serve pricing returns with a credit meter: Free $0 (50 credits), Pro $25/mo (200 builder credits), Business $79 per user/mo (500 builder credits), Enterprise custom.
• 2026-03 — Repack around workflow servers: Free (30 credits/mo), Pro $25/mo, Business drops from $79 per user to $50 flat, Enterprise custom.
• 2026-05-07 — Pivot: ‘Introducing Vellum: Your own Personal Intelligence’, with $25M announced from Dharmesh Shah, Arash Ferdowsi, Rebel Fund, and YC.
• 2026-06-10 — Current configurator pricing captured: free Base, Pro from ~$50/mo ($10 fee + machine + storage), pay-as-you-go credits at $1 = 1 credit, free MIT self-hosting.
• 2026-07-29 — Pro splits into three named presets — Mighty $30/mo, Super $100/mo, Ultra $200/mo — each bundling a machine tier, storage, and monthly credits, alongside a Custom a-la-carte build still from $50/mo; base storage rises 4 GiB to 6 GiB, and new Custom storage tops out at 120 GiB, with the former 250/500 GiB tiers grandfathered to existing subscribers only.
• 2026-08-28 — Custom’s optional recurring “credit bundle” is renamed “usage bundle” and its menu narrows from five flexible amounts ($10/$25/$50/$100/$200/mo) to exactly the three preset-matched tiers (Mighty $25, Super $45, Ultra $115/mo); subscribers already on a bundle at another amount are grandfathered. Docs also newly disclose that included usage does not roll over between billing cycles, purchased pay-as-you-go credits expire 12 months after purchase, and add a new Spending controls subsection (daily credit limit, low-balance alert). Headline preset prices are unchanged.

What’s unique : Vellum’s distinctive pricing mechanics

1. At-cost credit pass-through. Vellum states that when you spend $1 of credits on LLM tokens, the full dollar goes to the model provider — no inference markup. The margin lives in the $10 platform fee, machine tiers, and storage. That is close to unique among assistant products, which typically resell inference at 2-5x or hide it in a flat subscription.

2. Named presets bolted onto a configurator, not a retreat to plan tiers. Since July 29, 2026, Pro leads with three named starting points — Mighty ($30), Super ($100), Ultra ($200) — but each is just a pre-set combination of the same machine, storage, and credit dials Custom exposes individually, and editing any single dial on a preset silently converts the whole plan back to Custom. Vellum gets the fast-decision benefit of a Good/Better/Best ladder for new users without giving up the configurator’s resource-level granularity for power users. On August 28, 2026, that coupling got tighter still: Custom’s recurring credit dial — renamed “usage bundle” — was cut from five arbitrary amounts to exactly the three dollar figures already baked into Mighty/Super/Ultra, so the same three numbers now anchor pricing everywhere a buyer looks, whether they start from a preset or assemble Custom from scratch.

3. Open-source escape hatch on every axis. The assistant is MIT-licensed: self-host and the platform fee, machine charges, and storage tiers all disappear — you bring your own hardware and API keys. The hosted product competes with its own free version purely on convenience, which disciplines every hosted price.

Strengths & weaknesses

Billing UX : Vellum billing controls and transparency

• Billing controls — No credit card to start on Base. All plan and billing changes live in Settings → Billing: pick a preset (Mighty/Super/Ultra) or start a Custom build via Stripe Checkout, active immediately. Upgrading a tier bills the price difference right away (a bigger usage bundle unlocks extra usage as soon as that charge succeeds); a smaller machine applies immediately with the difference credited to the next invoice, while a smaller usage bundle waits until the next renewal — either way, no mid-cycle cash refunds. Editing any single tier on a preset auto-converts the plan to Custom. Cancellation is “Downgrade to Base” from the plans page, effective at the end of the current billing period.
• Usage visibility — The Billing screen shows a live Credit Balance plus controls for adding credits, Auto-Reload, a daily credit limit, and low-balance alerts; a separate Usage tab breaks down credit spend by day, model, and named “Actions” (main agent replies, memory consolidation, heartbeats, conversation summarization, etc.) so users can see and tune down background burn.
• Credit controls — Add Credits tops up $10-$100 per purchase via Stripe; Auto-Reload exposes three user-set values — a reload-below threshold ($1-$100, default $100), a reload amount ($10-$500, default $10), and an optional monthly spending cap ($25-$10,000) that pauses further auto top-ups for the month. Custom-plan users can alternatively add a recurring usage bundle (Mighty $25 / Super $45 / Ultra $115 per month — renamed from “credit bundle” and narrowed from five flexible amounts as of Aug 2026) as a subscription line instead of manual top-ups.
• Spending controls — A daily credit limit (optional, UTC-day boundary) pauses paid usage until the next day and emails the account owner when hit; a low-balance alert emails when the balance drops under a user-set threshold. Included usage resets each billing cycle with no rollover; purchased pay-as-you-go credits expire 12 months after purchase.
• Payment options — Self-serve card checkout via Stripe for Pro packages, Custom builds, and credit purchases. Self-hosters pay nothing to Vellum and bring their own model API keys.

Strategic wins : Why Vellum’s pricing decisions worked

1. Zero-margin inference as a trust wedge

A personal assistant sees your email, calendar, files, and memory — trust is the entire purchase decision. Pricing inference at cost removes the suspicion that the vendor profits from your assistant thinking more, and aligns Vellum’s revenue with resources users can verify (vCPU, GiB). It also future-proofs the model against token-price deflation. See choosing the right usage metric.

2. Selling compute and convenience, not capability

With the assistant MIT-licensed, Vellum cannot charge for the software — so the configurator monetizes exactly what self-hosters would otherwise provide themselves: a machine, a disk, an always-on subdomain. That is an honest split that converts the open-source funnel instead of fighting it. Related: how AI companies structure pricing.

3. A full pricing reset to match a full pivot

Rather than dragging enterprise SKUs into a consumer product, Vellum rebuilt pricing from zero — free Base, one configurable paid plan, one usage meter. After two years of quote-only enterprise pricing, the relaunch shipped with every price public and self-serve, which is what the prosumer audience it now courts expects. See usage-based pricing strategy and outcome-based pricing trends.

4. Presets layered on, not swapped in for, the configurator

On July 29, 2026, Pro gained three named presets — Mighty ($30), Super ($100), Ultra ($200) — sitting on top of the same configurator that previously required assembling a platform fee, machine tier, and storage tier from scratch. Editing a single dial on any preset still drops you into the pre-existing Custom builder, so power users lose nothing; new users just get a faster default. Most usage-priced products either keep a bare calculator indefinitely or replace it wholesale with fixed tiers — threading both at once is the less common move. See the shift toward usage-based packaging.

A month later, on August 28, 2026, Vellum tightened that same coupling: Custom’s optional recurring credit top-up — renamed “usage bundle” — was narrowed from five arbitrary amounts to the exact three dollar figures already bundled into Mighty, Super, and Ultra. The presets and the configurator now literally share a price list for the one dial that used to be independent, trading a little Custom flexibility for a pricing structure that reads the same no matter which path a buyer starts from.

Areas to improve : Gaps in Vellum’s pricing approach

1. Make idle credit burn predictable

Background operations — memory consolidation, summarization, hourly heartbeats — consume credits even when you are not using the assistant, and the pricing page does not quantify them. A published ‘typical idle cost per day’ figure (or an idle-cost estimate in the configurator) would prevent the classic first-bill surprise. See bill shock and cost unpredictability.

2. Extend the folded-in platform fee to Custom too

As of July 29, 2026, Super and Ultra already bundle the $10/mo platform fee into one flat price — exactly the fix this section used to call for. Mighty skips the fee outright, and Custom still itemizes it as a separate $10/mo line next to the machine and storage tiers it always had. Applying the same folded-fee treatment to Custom, or documenting why hand-assembled builds are priced differently from the presets, would remove the last place a Vellum bill still shows an unexplained flat charge.

3. Close out the enterprise era publicly

Wayback snapshots show paying self-serve customers on the agent platform as late as March 2026, yet no public migration or sunset notice exists for the enterprise product. For a company now selling trust to individuals, documenting how it treated its previous customer base is pricing-adjacent reputation work left undone.

4. Fix the terminology drift between the marketing FAQ and the docs

The August 28, 2026 rename from “credit bundle” to “usage bundle” reached the docs pricing page, the Custom builder, and the plan cards (which now read “Mighty/Super/Ultra Usage included” instead of dollar figures) — but the main pricing page’s own on-page FAQ (“Do plans include credits?”) still answers “Pro packages include a monthly credit bundle” as of this capture. A buyer scrolling from the plan cards down to the FAQ on the same page hits two different names for the same optional add-on.

Key takeaways

1. A pivot deserves a pricing reset. Vellum did not retrofit enterprise SKUs onto a consumer product — it rebuilt the model around the new buyer: free Base, one configurable Pro, one meter.
2. At-cost pass-through is a positioning weapon. Taking zero margin on inference reframes the vendor as being on the user’s side, and shifts monetization to resources buyers can verify.
3. Open source forces honest pricing. When self-hosting is free and MIT-licensed, hosted pricing must map to real costs (compute, storage, uptime) — capability rents are off the table.
4. Configurators and named tiers aren’t mutually exclusive. Vellum spent over a year selling Pro purely as an assemble-your-own configurator, then on July 29, 2026 added three named presets (Mighty/Super/Ultra) that are just pre-set configurator outputs — proof that a fast default for most buyers and full resource-level control for power users can coexist in the same product.
5. Pricing whiplash has a memory. Public tiers, then two years of quotes, then credits, then a pivot — each reset was individually defensible, but the sequence is a trust tax Vellum now has to pay down with transparency.

UBP implications

1. Split the meter from the subscription — even inside a bundled preset. Vellum charges fixed monthly for capacity (machine, storage) and pure usage for consumption (credits); the July 2026 presets fold a credit allowance into one sticker price for convenience, but usage beyond that allowance still falls back to the same pay-as-you-go credit meter, keeping the underlying separation intact.
2. Pass-through pricing only works if you monetize something else. Zero-margin inference is sustainable here because compute tiers and the platform fee carry the margin; copying the pass-through without a margin carrier is a race to zero.
3. Spending caps belong in the product, not the FAQ. Auto-reload bounded by a user-set monthly cap ($25-$10,000) is the kind of guardrail every credit-based AI product should ship on day one. See usage-based pricing strategy.

Sources

• Vellum pricing page — live capture, independently re-verified (accessed 2026-08-28)
• Vellum pricing docs — credits, machine tiers, auto-reload — independently re-verified (accessed 2026-08-28)
• Vellum Assistant v0.10.12 release notes — “Billing, simplified” — independent off-domain confirmation of Mighty $30/Super $100/Ultra $200 preset pricing, still current through v0.11.7 (accessed 2026-08-28)
• Introducing Vellum: Your own Personal Intelligence — 2026-05-07 (accessed 2026-06-10)
• vellum-ai/vellum-assistant on GitHub (MIT) (accessed 2026-06-10)
• Announcing our $20M Series A — Vellum, 2025-07-10 (accessed 2026-06-10)
• Y Combinator — Vellum (W23) (accessed 2026-06-10)
• Wayback Machine snapshots: vellum.ai/pricing — 2023-06, 2024-08, 2025-07, 2025-12, 2026-03 (accessed 2026-06-10)

Bottom line

Vellum is the corpus’s clearest example of a pricing model rebuilt to match a pivot: the YC W23 enterprise LLM-platform company ($5M seed, $20M Series A) relaunched in May 2026 as an open-source personal AI assistant with $25M announced, and replaced years of quote-heavy enterprise pricing with a fully public configurator — free Base, Pro starting at $30/mo across three named presets (Mighty/Super/Ultra) or a self-assembled Custom build from $50/mo, and pay-as-you-go credits passed through to model providers at cost. An August 28, 2026 tweak tightened the model further, narrowing Custom’s optional recurring credit top-up (renamed “usage bundle”) to the exact three amounts already baked into the presets — no headline price moved. The MIT self-host path keeps hosted prices honest; the open questions are idle credit burn, the platform fee’s inconsistent treatment across presets, the undocumented fate of its enterprise customers, and now a “credit bundle” vs. “usage bundle” naming split between its own FAQ and its docs. Browse the pricing blueprint for more fully-researched company profiles.

Want to compare Vellum against other AI assistant and agent companies? Browse the pricing blueprint.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Custom plan's credit bundle renamed usage bundle, narrowed to 3 tiers

Aug 2026

Vellum renamed the Custom plan's optional recurring credit top-up from 'credit bundle' to 'usage bundle' and cut its menu from five flexible amounts ($10-$200/mo) to the three amounts already bundled into the Mighty/Super/Ultra presets ($25/$45/$115/mo), grandfathering existing subscribers on other amounts. Docs also newly disclose that included usage doesn't roll over between billing cycles, purchased credits expire 12 months after purchase, and add a new 'Spending controls' subsection (daily credit limit, low-balance alert).

Pro repacks into Mighty / Super / Ultra presets plus Custom

Jul 2026

The single ~$50/mo Pro configurator splits into three named presets bundling machine, storage, and credits — Mighty $30/mo, Super $100/mo, Ultra $200/mo — plus a Custom a-la-carte build still from $50/mo. Base storage rises 4 GiB to 6 GiB; new Custom storage tops out at 120 GiB, with the former 250/500 GiB tiers grandfathered to existing subscribers only.

Pivot pricing — personal AI assistant configurator

Jun 2026

After the May 7, 2026 'Personal Intelligence' relaunch ($25M announced), pricing becomes a configurator: free Base (Small machine, 4 GiB storage), Pro from ~$50/mo ($10 platform fee + machine tier +$35 to +$125 + storage +$5 to +$120), credits pay-as-you-go at $1 = 1 credit, and free MIT self-hosting.

Agent-platform repack: Business drops to $50 flat

Mar 2026

Tiers reframed around workflow servers and concurrent runs: Free $0 (30 credits/mo), Pro $25/mo (100 credits), Business $50/mo flat (12 concurrent runs), Enterprise custom. Last enterprise-era pricing before the pivot.

Self-serve credit pricing returns

Dec 2025

Public prices return with a credit meter: Free $0 (50 credits), Pro $25/mo (200 builder credits), Business $79/user/mo (up to 5 users, 500 builder credits), plus custom Enterprise — pricing now anchored on hosted agent apps.

$20M Series A; pricing stays gated

Jul 2025

Days after announcing a $20M Series A led by Leaders Fund (July 10, 2025), the page still shows quote-only Startup / Pro / Enterprise plans with 'Book a Demo' CTAs and no public prices.

Prices pulled — quote-only 'Request Demo' era

Aug 2024

All published prices removed. Pricing page restructured to Growth / Pro / Enterprise with 'Request Demo' CTAs and no dollar figures — a fully sales-led motion aimed at enterprise AI teams.

Launch-era public tiers for the LLM dev platform

Jun 2023

Five public tiers with a 7-day free trial: Free $0, Starter $49/user/mo, Growth $249/mo, Pro $699/mo, and custom Enterprise — limits set by LLM production requests per week, deployments, and seats.