Devgraph
Devgraph builds a live ontology of your code, infrastructure, tickets and teams, then lets engineers and AI agents query it in plain language. Native MCP support, bring-your-own model, and connectors for GitHub, Jira, Argo CD and Grafana.
What is Devgraph?
Devgraph is an ontology engine for software organisations, published by Arctir, Inc, a Delaware corporation that presents it as its core product. Its premise is that the knowledge an engineering team needs is scattered across a dozen systems, and that neither people nor AI models can reason about it while it stays that way. Devgraph connects to those systems, discovers what is inside them, and assembles the result into one queryable graph of code, infrastructure, tickets and people.
The connectors are called molecules. Nine are documented: GitHub, GitLab, Argo CD, Vercel, LDAP, Grafana, Jira, Docker and FOSSA. Each authenticates against an external API, creates entities and typed relations such as HOSTED_BY, USES, MEMBER_OF and DEPLOYS, then reconciles on a schedule you set, so the graph tracks the present rather than describing a past state. Field-based selectors let a relation link entities that different molecules created, which is how an Argo CD application ends up connected to the GitHub repository it deploys. Entities themselves are declared in Kubernetes-style YAML, under an entities.devgraph.ai/v1 API version.
What that buys is answers. The documentation gives examples: which services depend on this repository, which deployments the platform team owns, which Docker images are running in production, which Argo applications use a given repo. It extends to actions too, such as opening issues on every repository affected by a CVE, or redeploying the Analytics services.
Two things set it apart. The first is native Model Context Protocol support: AI agents query Devgraph before acting, and every pricing tier includes a quota of MCP servers. The second is that Devgraph ships no model of its own. You bring OpenAI, Anthropic, xAI, Ollama or something you host yourself, and self-hosted and air-gapped deployments are advertised.
Discovery runs either inside the platform through the web interface, or locally from a YAML file suited to CI/CD pipelines. A REST API of roughly a hundred endpoints covers entities, relations, environments, MCP endpoints, model providers, prompts, tokens and chat completions. The public documentation, on docs.devgraph.ai, is labelled Version: Beta.
What it does
- Build a live ontology of your code, infrastructure and teams
- Query the whole stack in plain language from a single place
- See which services a change will affect before you deploy it
- Find out who owns a service and which teams need warning
- Expose the ontology to AI agents through MCP servers
- Run it all on the LLM of your choice, hosted wherever you want
- Let agents act: open tickets, redeploy services, refresh documentation
When to use Devgraph / When not to
A quick filter to help you decide if Devgraph is the right fit.
When to use Devgraph
- Platform engineering and SRE teams that need a live map of how services depend on each other
- Organisations grounding AI agents in their real systems through the Model Context Protocol
- Engineering groups that onboard newcomers often and want ownership answers in hours rather than weeks
- Teams for whom impact analysis before a deployment is a routine, non-negotiable step
- Companies that insist on choosing their own LLM, including self-hosted or air-gapped setups
When not to use Devgraph
- Users outside the United States: the terms declare the service intended solely for US use
- Organisations bound by the GDPR, since the vendor states it does not direct services to the UK or the EEA
- Solo developers and very small teams, given a 99 USD entry tier capped at two users and 500 entities
- Anyone shopping for a code-generation assistant, because Devgraph reads systems rather than writing code
- Teams needing a permanently free tier, as only 14-day and 30-day trials are on offer
How to use Devgraph
A typical end-to-end flow, from setup to results.
- Create an account on the Devgraph dashboard; sign-in runs through Clerk, with Google or GitHub OAuth available
- Create your first environment, which will hold everything that follows
- Decide how discovery should run: inside the platform through the web interface, or locally from a YAML file
- Pick a first molecule to connect, such as GitHub, GitLab, Argo CD or Grafana
- Issue the credential that molecule expects: a personal access token or GitHub App, a GitLab private token, an Argo CD auth token, a Grafana API key, or an LDAP bind DN
- Pass those secrets through environment variables instead of writing them into the configuration
- Set the shared options: provider name, type, namespace, and the every interval that drives reconciliation
- Narrow the scope with selectors, filtering by organisation, repository name pattern, tags or folders
- Run discovery and let Devgraph create the entities and typed relations in your graph
- Query the ontology in plain language, or attach MCP endpoints so your agents can read it too
Pros & Cons
Pros
- The choice of LLM stays with the customer, self-hosted and air-gapped deployments included
- Native MCP support plugs the ontology straight into AI agents you already run
- A broad REST API and a detailed public documentation site
- Connectors span code, deployment, observability, issue tracking and identity
- Discovery can run locally from YAML, so it can be versioned and driven from CI/CD
- The terms leave ownership of Customer Data explicitly with the customer
- Entry and mid-tier prices are published openly, with a free trial on all three plans
Cons
- The terms declare the service intended solely for the United States, and the privacy policy excludes the UK and the EEA
- No GDPR compliance claimed, no Article 27 representative, no DPO, and no data processing agreement from the vendor
- No published security certification: neither SOC 2, nor ISO 27001, nor HIPAA
- No postal address anywhere, neither on devgraph.ai nor on Arctir's own about page
- No documented way to exclude your data from model training
- No free plan, and the 99 USD entry tier is capped at two users, 500 entities, one ontology and one environment
- Payments are non-refundable, cancellation must land at least a day early, and exports close five days after the term
Pricing & Plans
There is no permanently free plan. The lowest paid entry point is the Liftoff tier at 99 USD per month, billed in United States dollars and preceded by a 14-day free trial. Crew follows at 499 USD per month on the same trial terms, and Enterprise is quoted on request with a 30-day trial. A monthly/annual toggle appears on the pricing page, but no annual amount is rendered in the page itself and none could therefore be recorded.
- 1 ontology
- 1 environment
- up to 500 entities
- up to 5 MCP servers
- up to 5 discovery providers
- up to 2 users
- bring your own model
- 3 business day response time
- 1 ontology
- 1 environment
- up to 2
- 500 entities
- up to 10 MCP servers
- up to 10 discovery providers
- up to 25 users
- bring your own model
- unlimited ontologies
- environments
- entities
- MCP servers
- discovery providers and users
- bring your own model
- automatic backups
- on-premise deployment options
Data, GDPR & hosting
A consolidated view of how Devgraph handles your data.
GDPR overview
There is no GDPR implementation to report, and the privacy policy says as much in plain terms. Arctir states that it does not direct products or services to residents of the United Kingdom or the European Economic Area, and that it does not meet the CCPA definition of a business. The terms add that the service is operated from the United States and intended solely for US use. No Article 27 representative is designated, no data protection officer is named, and no legal basis or transfer mechanism is described. Arctir offers no data processing agreement of its own; the only DPA linked belongs to Clerk, a third-party processor. The rights actually offered are contractual: decline to provide data, manage or delete your account through the Clerk-powered settings or by email, and manage payment methods in the Stripe portal. Effective date: 3 November 2025.
Who owns the data?
Under the terms of use, you keep all right, title and interest in your Customer Data, intellectual property included. Arctir receives only a limited, non-exclusive, worldwide, royalty-free licence, for the duration of the contract and solely to deliver the service. You additionally authorise Arctir to anonymise your Customer Data and aggregate it with other customers' data, in a form that does not identify you, in order to develop its products. Usage Data also belongs to you, yet Arctir may aggregate and disclose it non-identifiably, both during and after the term. Feedback you send is licensed to Arctir perpetually and irrevocably.
Reuse rights
Your Customer Data stays yours to reuse as you see fit: no permission from Arctir is required, and you can export it from the application at any point while the subscription runs. Arctir is not obliged to back it up, so keeping your own copies is explicitly your responsibility, and the export window closes five days after the term ends. Material supplied by Arctir is treated differently. Company Materials come with a limited licence to copy and use them for personal or internal business purposes only; redistributing or publicly displaying them elsewhere needs written consent. Reverse engineering, running the service as a bureau for third parties, and building a competing product are all forbidden.
Data retention & training
Hosting summary
The services run on Amazon Web Services, named in the privacy policy as the hosting provider. AWS automatically logs technical data such as IP address, browser type, request timestamps and server logs, acting as a processor on Arctir's instructions. No hosting region and no storage country are stated anywhere on the site, so the jurisdiction in which customer data physically rests is simply not published. What is stated is operational: the terms declare the service operated from the United States and intended solely for US use, and Arctir is a Delaware corporation. The domain resolves to 76.76.21.21, an anycast node geolocated in the United States on Amazon's AS16509, which describes the front end rather than the storage. Enterprise customers can sidestep the question altogether: on-premise deployment options are listed on that plan, and the homepage advertises self-hosted and air-gapped deployments that keep both data and models on the customer's own infrastructure.
Where Devgraph works
Country-level availability.
Available in
Not available in
Things to keep in mind
Risks and trade-offs to weigh before adopting Devgraph.
- Automatic renewal, with a cancellation window that closes a single day before the period ends
- Payments are non-refundable, so a badly sized plan is simply money spent
- Customer Data stays retrievable for only five days after the term, and Arctir is not obliged to back it up
- You grant Arctir a licence to anonymise and aggregate your Customer Data, and Usage Data remains exploitable after the contract ends, with no stated time limit
- A live map of who owns what can quietly become a surveillance tool if managers read it as a record of individual activity rather than of systems
- Trusting graph answers without ever checking them erodes the mental model of the estate that engineers used to carry themselves, which is exactly what you need when the graph is wrong or down
- The terms can be changed without notice, and Google Analytics and interest-based advertising run while Do Not Track signals are ignored
Setup & Integrations
Technical difficulty
Moderate, and squarely aimed at technical teams. Account creation is immediate through Clerk, with Google or GitHub sign-in. Platform discovery, configured in the web interface, is presented as the easiest route. Local discovery demands a YAML file, an API credential per system - GitHub token or App, GitLab private token, Argo CD auth token, Grafana API key, LDAP bind DN - and a working grasp of selectors, namespaces, typed relations and reconciliation intervals. Advanced options such as on-premise, air-gapped or custom providers assume an infrastructure team. A non-engineer will not set this up alone.
Deployment
Integrations
Behind Devgraph
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Devgraph actually do?
Which tools does it connect to?
Which AI model does it use?
Is there an API?
Does it work with AI agents?
How much does it cost?
Can I host it myself?
Is the service available in Europe?
Who owns the data, and how long is it kept?
Should you pick Devgraph?
Devgraph is a focused tool that does not pretend otherwise. It sets out to give engineering organisations, and the AI agents they are starting to deploy, a single accurate picture of how their systems fit together, and it goes about that with a graph built from the tools teams already run rather than from documentation someone has to remember to update.
Its strongest cards are the live ontology itself, native Model Context Protocol support, and the decision to supply no model at all. Letting customers bring OpenAI, Anthropic, xAI, Ollama or a self-hosted model, and offering air-gapped deployments, answers the objection that usually stops this kind of product at the security review. The REST API is broad, the documentation is public and detailed, and discovery can be driven from a YAML file in CI/CD rather than clicked together by hand.
The reservations are mostly about the company rather than the software. The vendor declares the service intended for the United States alone and explicitly excludes the United Kingdom and the European Economic Area, which puts it out of reach of any organisation with GDPR obligations. No security certification is published, no postal address appears anywhere, no data processing agreement is offered, and no mechanism exists to exclude your data from model training. The documentation still carries a Beta label, and the contract terms are tight: non-refundable payments, a cancellation window closing a day before renewal, and five days to retrieve your data once the term is over.
For a US platform engineering or SRE team with a real dependency-mapping problem, the free trial on all three tiers makes it cheap to find out whether the graph earns its 99 USD. Everyone else should read the terms first.
- Choosing a selection results in a full page refresh.
- Opens in a new window.