MatterAI
MatterAI is an engineering platform pairing its own Axon reasoning models with an agentic coding IDE, a terminal CLI and automated pull-request reviews, promising frontier-class results at roughly 70% lower token cost. Free tiers cover solo developers.
What is MatterAI?
MatterAI is the product line of Gravity Cloud Services, Inc., a Delaware company that describes itself as a vertical AI lab and sells one idea: frontier-grade reasoning you can actually afford to run. Everything sits on the Axon model family — Lumen 4, the new flagship, plus Eido 3 Pro, Mini and Flash. All four carry a one-million-token context window and native text-and-vision input, and all four are reachable through an OpenAI-compatible endpoint, so an existing client library needs only a new base URL and key.
Three products share that engine. Orbital is the coding agent: its own IDE plus extensions for VS Code, Cursor and Windsurf, with an agent mode, a planning mode, full codebase indexing and tab completion, distributed as a macOS disk image. OrbCode is the terminal counterpart, an MIT-licensed CLI installed from npm that needs Node.js 20 or later; it renders streaming markdown and live reasoning in a native terminal UI, shows real diffs before every file edit, asks for approval on edits and commands, refuses to auto-approve dangerous ones, runs headless for CI, loads SKILL.md files and AGENTS.md project memory, and speaks to MCP servers over stdio, HTTP and SSE. AI Code Reviews closes the loop on GitHub, GitLab and Bitbucket: agentic review of each pull request, a confidence score on every finding, automated fixes delivered as a separate PR, and analytics covering DORA metrics, cycle time and review participation. A dedicated MCP server triggers those same reviews from Cursor or Windsurf.
The reasoning approach has a name, Panoptic Thinking: a persistent execution graph that keeps the whole repository in working memory, an internal question-and-answer map, and a plan re-synthesised before every action rather than a single linear pass. The company also markets the models beyond code — support agents, internal workflow automation, a memory engine, research agents, analytics copilots and document extraction.
Enterprise controls are unusually deep for the price: VPC and on-premises deployment, SSO through Okta, Azure AD or Google Workspace, bring-your-own-model, LLM observability, and a data-residency region chosen between the United States, the European Union and Singapore. The vendor claims SOC 2 Type II certification, 11,000 or more engineers, over a billion tokens a day and 1.2 million pull requests reviewed.
What it does
- Review pull requests automatically and post context-aware comments with a confidence score
- Generate fix commits and open them as an isolated Fix PR
- Detect security flaws before commit, including SQL injection, XSS and broken authentication
- Write and refactor code in the Orbital IDE or through VS Code, Cursor and Windsurf extensions
- Run agentic coding sessions in the terminal, with multi-file edits and approved shell commands
- Serve the Axon models to your own agents through an OpenAI-compatible API
- Report team analytics, from DORA metrics to review response time and merge rate
When to use MatterAI / When not to
A quick filter to help you decide if MatterAI is the right fit.
When to use MatterAI
- Engineering teams that want every pull request reviewed automatically on GitHub, GitLab or Bitbucket
- Terminal-first developers who want an agentic coding CLI they can inspect, since OrbCode ships under an MIT licence
- Teams running agents in production and looking to cut their inference bill, with input tokens billed between 0.50 and 5 US dollars per million
- Regulated organisations that need VPC or on-premises deployment, SSO and a choice of data-residency region
- Maintainers of public open-source repositories, who get the full code-review product at no cost
When not to use MatterAI
- Anyone expecting a mobile app: the only desktop client is a macOS disk image, with no iOS or Android release
- Non-technical roles such as marketing, design or writing, since every surface assumes a Git repository and a development environment
- Teams working in a language other than English, as the interface is English-only
- Windows and Linux users who want the standalone Orbital desktop app rather than an editor extension
- Buyers who need independently audited benchmark leadership, because the vendor's own charts place Axon behind rival models on most tests
How to use MatterAI
A typical end-to-end flow, from setup to results.
- Create an account on the MatterAI web app, using the Get Started or Get API Key button
- For API work, generate a key and point any OpenAI-compatible client at the inference endpoint
- For automated reviews, connect a GitHub, GitLab or Bitbucket repository
- Open a pull request and let the agent comment, score its findings and propose a Fix PR
- For the visual editor, download the macOS build of Orbital or install the VS Code, Cursor or Windsurf extension
- For the terminal, install the OrbCode package globally from npm on Node.js 20 or later
- Run the login command once and authenticate in the browser
- Change into a project directory and start OrbCode there, since that directory becomes the workspace
- Approve or reject each file edit and each shell command as the agent proposes them
- For an enterprise rollout, book a demo, then configure SSO, the residency region and VPC or on-premises hosting
Pros & Cons
Pros
- Inference priced far below frontier rivals, from 0.50 to 5 US dollars per million input tokens
- A one-million-token context window across the whole model family, at one flat price for the full window
- The full chain covered by one vendor: terminal, editor and pull request, with no tools to stitch together
- Flat pricing on code reviews, 49 US dollars a month for up to ten seats, where rivals bill per seat
- OrbCode is MIT-licensed and its hooks are Claude Code compatible, so that layer is not a lock-in
- Enterprise options rare at this price: VPC or on-premises deployment, bring-your-own-model and a choice of residency region
- A genuinely free path, with a permanent free tier and full free access for public open-source repositories
Cons
- The pricing page renders nothing without JavaScript, so the amounts are invisible to anyone reading the raw page
- The homepage promises code is never used to train the base models, while the Professional Terms allow training unless you opt out
- The subprocessor list the data processing agreement points to returns a 404, so you cannot see who actually handles your data
- Two different legal entities appear across the contracts, Gravity Cloud Services, Inc. and MatterAI Services Inc.
- The Orbital Chat link in the site navigation points at a subdomain that does not respond
- The vendor's own benchmark charts place Axon behind competing models on most of the tests shown
- No mobile app, and the standalone desktop build is macOS only
Pricing & Plans
A permanent free plan is available on two fronts: the Developer tier for code reviews at no cost, and an Orbital tier advertised as free forever. Paid plans also come with a 14-day free trial requiring no card. The cheapest paid entry point across the whole catalogue is Orbital Pro at 10.00 USD per month; on the code-review side, the entry plan is Startup at 49.00 USD per month for up to ten users. API inference is billed separately, per million tokens. Public open-source projects pay nothing.
- 15 reviews and 25 summaries a month
- 1 user
- community support
- 250 reviews and 500 summaries a month
- up to 10 users
- AI Security Scan
- AI Release Notes
- Discord and email support
- 14-day free trial
- 2
- 000 reviews and 4
- 000 summaries a month
- up to 50 users
- PR Analytics
- Custom Rulesets
- AI Memories
- AI Documentation
- bring your own model
- unlimited PRs and users
- LLM Observability
- SSO
- dedicated support
- self-hosting option
- Axon Eido 3 Flash
- Mini and Pro
- 200k context
- limited usage
- advertised as free forever
- 200k context
- 5.5x the free tier's usage
- adds Axon Lumen 4
- 400k context
- twice the Pro usage (highlighted plan)
- 400k context
- five times the Pro usage
- Axon Lumen 4 at 5.00 USD in and 25.00 USD out
- Eido 3 Pro at 3.00 and 9.00
- Eido 3 Mini at 1.00 and 3.00
- Eido 3 Flash at 0.50 and 1.50
Data, GDPR & hosting
A consolidated view of how MatterAI handles your data.
GDPR overview
Implementation is contractual and fairly complete. MatterAI publishes a full data processing agreement at /dpa, including the EU standard contractual clauses with their annexes, and the Professional Terms require a signed DPA before any personal data governed by the GDPR is submitted. The privacy policy, effective 8 August 2025, sets out a legal-basis table for the EEA and the UK (contract, legitimate interests, consent, legal obligation), lists access, rectification, erasure, restriction, objection and portability rights, and acknowledges international transfers under standard contractual clauses. An eu-west residency region is offered. Two gaps stand out: no Article 27 representative is designated, and the subprocessor list the DPA points to returns a 404. The site never states in so many words that it is GDPR compliant.
Who owns the data?
The Professional Terms keep three categories apart. Customer Content stays with the customer: MatterAI may copy, display, modify and use it only as needed to provide and maintain the product, and the customer remains responsible for its accuracy. Feedback is different — it is given "as is" and MatterAI may use it freely, without restriction or obligation. Usage Data may be collected and analysed at will, but disclosed to third parties only in aggregated, non-identifying form. MatterAI retains every right in the product itself. Note that the published contracts name two Delaware entities: Gravity Cloud Services, Inc. in the footer and privacy policy, MatterAI Services Inc. in the data processing agreement.
Reuse rights
Customers keep their own content and may reuse it without asking permission; the restrictions in the Professional Terms bear on the product, not on customer data. Reverse engineering, resale, sublicensing, unauthorised security testing and building a competing service are all forbidden. Customer Content is deleted within 60 days of a customer request. Training is the point to read closely: the terms allow MatterAI to train its models on Materials unless the account owner opts out in the account settings, and even after opting out, submitted Feedback and content flagged for safety review remain usable for training. One component escapes all of this — the OrbCode CLI is published under an MIT licence on GitHub and can be reused freely.
Data retention & training
Hosting summary
Customers choose one of three residency regions: the United States (us-east), the European Union (eu-west) or Singapore (ap-southeast), with the site stating that data never leaves the chosen region. For reviews, the repository is cloned into memory, analysed and then purged, with no persistent storage of proprietary code claimed; secure multi-tenant storage covers generation and review in progress. The privacy policy acknowledges international transfers, including to the United States, under standard contractual clauses. The data processing agreement adds a nuance worth knowing: where data sits in the EEA datacentre, only services whose subprocessor has EEA facilities are enabled by default, and using third-party integrations is expected to move data out of the EEA. VPC and on-premises deployment are offered for full isolation. One gap: the subprocessor list referenced by the agreement is not reachable.
Things to keep in mind
Risks and trade-offs to weigh before adopting MatterAI.
- Marketing and contract disagree on training: read the Professional Terms, not the homepage, before submitting proprietary code
- Opting out of training does not cover everything, since submitted feedback and safety-flagged content remain usable
- The subprocessor list is unreachable, so you cannot verify which third parties touch your data
- Your source code leaves your infrastructure for analysis, an intellectual-property exposure to weigh against the VPC or on-premises option
- Automated fix commits merged without a human read can introduce regressions where test coverage is thin
- Claims such as detecting and fixing 95% of bugs, or a 40% bug reduction, come from the vendor and are not independently audited
- Leaning entirely on agentic review can erode a team's own reviewing skill over time
Setup & Integrations
Technical difficulty
Low to moderate, but technical throughout: every path assumes a Git repository and a development environment. The quickest route is to create an account and connect a GitHub, GitLab or Bitbucket repository, after which reviews run on their own. OrbCode is a global npm install on Node.js 20 or later, followed by a browser sign-in — three commands, as the site puts it. Orbital means downloading a macOS build or installing an editor extension. API use requires no rewriting, since the endpoint is OpenAI-compatible. Enterprise deployments with VPC, on-premises hosting, SSO or bring-your-own-model are an infrastructure project and go through sales.
Deployment
Integrations
Supported languages
Behind MatterAI
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement MatterAI.
Frequently asked questions
Which models does MatterAI offer?
Is there an API?
Is there a free plan?
What does the cheapest paid plan cost?
Which Git platforms and editors are supported?
Will my code be used to train the models?
Where is my data hosted?
How long is data kept?
Is a DPA available, and are there certifications?
Is there an offer for open-source projects?
Should you pick MatterAI?
MatterAI is an unusually complete stack for a young company. One engine, the Axon model family, feeds a terminal CLI, an editor agent and automated pull-request reviews, and the same models are sold raw through an OpenAI-compatible API. The pricing is the argument: input tokens from 0.50 to 5 US dollars per million, a flat 49 dollars a month covering ten seats on code reviews, a paid entry point at 10 dollars, a permanent free tier and full free access for public open-source repositories. Enterprise controls that usually appear much higher up the market — VPC and on-premises deployment, bring-your-own-model, SSO, and a residency region chosen between the United States, the European Union and Singapore — are on the table here too.
Three things deserve care before committing. The training story is inconsistent: the homepage says your code is never used to train the base models, while the Professional Terms permit exactly that unless you opt out in your account settings. The subprocessor list required by the data processing agreement returns a 404, so the chain of processors cannot be inspected. And two separate Delaware entities appear across the documents, one in the footer and privacy policy, another as the party to the DPA. None of these is disqualifying, but each is worth a written answer from the vendor.
The product itself looks alive: the desktop build is served, the web app and documentation respond, and the OrbCode package was republished on the day of this review. Worth noting that the vendor's own benchmark charts place Axon behind several rivals on most tests — the pitch is value for money, not raw supremacy, and it should be read that way.
- Choosing a selection results in a full page refresh.
- Opens in a new window.