# Arsum — AI build studio scoping and delivering production AI applications, search systems and integrations.

> Source: The Agents Index — https://theagentsindex.com/arsum (structured, researched, re-verified)
> Facts last verified: 2026-10-03

Arsum is a US-registered AI systems engineering studio that scopes, builds and hands over production AI applications, intelligent search and retrieval systems, and AI integrations into existing products. Engagements cover architecture, data connections, acceptance testing, deployment and post-launch ownership.

| Fact | Value |
| --- | --- |
| Website | https://arsum.com |
| Price | from $5,000/engagement |
| Free tier | No |

No free or trial tier: work begins with a project brief and a first call that decides whether discovery, a scoped build, an integration, support or no project follows.

## What it does

**Writes code** · Yes — Custom application and back-end code is written by Arsum's engineers as part of a scoped build, covering interface, back-end and model integration together.

> Scope the interface, back-end, model integration and required system connections together.
> — https://arsum.com/services/ai-products-applications/ · "Build one useful product workflow first" section, "Build the application" step

**Debugs and fixes** · Partly — Support and incident response on a live AI system is one of four named engagement types, but it is scoped per contract and the vendor explicitly declines to promise a managed-operation service or response time.

> This page does not promise an established managed-operation service, a response-time guarantee, or a measured client outcome.
> — https://arsum.com/ai-integration-services/ · Closing paragraph below the "Make the next decision" step

**Writes tests** · Yes — Acceptance testing against representative examples and failure cases is part of the build scope, including permissions, invalid inputs and recovery paths.

> Use representative examples and failure cases to agree on acceptance. Include permissions, invalid inputs, approval steps, cost limits and recovery from unavailable services.
> — https://arsum.com/services/ai-products-applications/ · "Test the behavior" step

**Ships it** · Yes — Deployment is inside the engagement rather than left to the client, and the handover step names the deployment environment and rollback explicitly.

> Define the deployment environment, code and account ownership, documentation, rollback and support boundary.
> — https://arsum.com/services/ai-products-applications/ · "Hand over a usable system" step

**Done-for-you service** · Yes — Arsum is a delivery studio, not software you license: its people scope, build, deploy and then hand over the system, with support scoped separately.

> Arsum scopes and builds production AI applications, intelligent search systems, and AI integrations. We connect the data, test real cases, deploy the software, and define who owns it after launch.
> — https://arsum.com/ · Hero section, paragraph under the main heading

_Not established: Reviews code, Whole-codebase work_

## How much it does unattended

**Runs autonomously** · No — The delivered pattern deliberately keeps a person in the loop: the system prepares output and a named human approves release, which Arsum states as a design boundary.

> The new application automates that preparation, but it does not give the model ownership of the final release. A practitioner reviews the generated result before it becomes available to the customer.
> — https://arsum.com/case-studies/attuned-health/ · Section 04 'Human responsibility'

**Agent permissions** · Partly — Permission boundaries are a design decision Arsum settles per engagement rather than a fixed product default, deciding where approval is required before action.

> Decide whether the system may draft, recommend, or act, and which actions require approval.
> — https://arsum.com/ai-integration-services/ · "Map the data and action" step

## Which models it runs on

**Claude** · Partly — Anthropic models appear in the published implementation stack of one delivered client build, which shows Arsum works with Claude but is not a product-wide support statement.

> AI workflow OpenAI · Anthropic
> — https://arsum.com/case-studies/attuned-health/ · Section 03 'The implementation stack' block

**GPT** · Partly — OpenAI models are named in the same delivered stack, evidencing hands-on GPT work on a client engagement rather than a supported-model list across all projects.

> AI workflow OpenAI · Anthropic
> — https://arsum.com/case-studies/attuned-health/ · Section 03 'The implementation stack' block

**Model choice** · Partly — Model choice sits with Arsum's architecture work and is expected to change over time; two different providers appear in delivered work, so the client is not locked to one vendor.

> A production system must survive real data, permissions, exceptions, changing models, and real users.
> — https://arsum.com/ · 'AI systems that work beyond the demo' section

_Not established: Open models_

## Whose machine it runs on

**Self-hosted** · Partly — The system Arsum builds runs in an environment the client owns, with code and account ownership settled at handover, but there is no Arsum platform to self-host.

> Define the deployment environment, code and account ownership, documentation, rollback and support boundary.
> — https://arsum.com/services/ai-products-applications/ · "Hand over a usable system" step

_Not established: Open source_

## What it costs to run

**How it meters** · Yes — Pricing is per scoped engagement with a published target range, and third-party model, hosting and support spend is estimated on top rather than bundled.

> Our target for an initial scoped engagement is USD $5,000–$20,000, with longer projects possible. ... Model usage, hosting and ongoing support are estimated explicitly.
> — https://arsum.com/services/ai-products-applications/ · "Budget and delivery scope" section

**Free tier** · No — There is nothing to sign up for: engagement begins with a project brief and a call that decides whether any project happens at all.

> The first call identifies whether the next step is discovery, a scoped build, an integration, support, or no project.
> — https://arsum.com/ · "Who Arsum works with" section

**Bring your own key** · Partly — Inference and hosting are quoted as the client's own explicit costs rather than resold by Arsum, which implies client-held provider accounts, though keys are never discussed directly.

> Model usage, hosting and ongoing support are estimated explicitly.
> — https://arsum.com/services/ai-products-applications/ · 'Budget and delivery scope' section

## Buying it for a team

**A company can buy it** · Yes — The buyer is a company: Arsum addresses founders, product and technology leaders and asks for the person who can accept the work.

> Arsum works with founders, product leaders, CTOs, and functional owners who need an AI product or workflow to perform in the real world.
> — https://arsum.com/about/ · Intro paragraph under the page heading

**Admin controls** · Partly — Access control is designed into the systems Arsum builds, applying the user's own access rules before data reaches a result or the model, but no Arsum admin console is offered.

> Apply the user’s access rules before information enters a result or model context.
> — https://arsum.com/services/intelligent-search-data-systems/ · "Enforce access and uncertainty" step

**Audit log** · Partly — Audit records are named as something specified during an integration's exception planning, so they are scoped work rather than a standing feature a buyer gets.

> Specify retries, escalation, audit records, rollback, and what happens when a source or model is unavailable.
> — https://arsum.com/ai-integration-services/ · "Plan for exceptions" step

_Not established: Single sign-on_

## What happens to your code

**Data residency** · Partly — Arsum LLC is a United States entity in Wyoming and reserves the right to process data outside the country of collection, with no jurisdiction choice offered to buyers.

> Arsum and its service providers may process information in countries other than the one where it was collected. Where required, we use appropriate safeguards for international transfers.
> — https://arsum.com/privacy/ · Section 8, "International transfers"

**Getting out** · Yes — Exit is a defined deliverable: code and account ownership, documentation, rollback and the support boundary are settled at handover rather than left open.

> Define the deployment environment, code and account ownership, documentation, rollback and support boundary. Launch acceptance and later feature development need separate definitions.
> — https://arsum.com/services/ai-products-applications/ · "Hand over a usable system" step

_Not established: Opt out of training, Certifications_

## Provenance

Every fact above comes from the vendor. Nothing here is independently corroborated yet.

## Coverage elsewhere

None recorded. Every fact above comes from the vendor and is not independently corroborated yet.

## What changed

- 2026-10-03 — Capabilities · Commercial terms · Free tier · Price

## Correcting this record

Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Arsum, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth.

Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.