Questions about CodeZero and Vörr

CodeZero (also written Code0) is an AI engineering workspace that runs as a VS Code extension. It gives developers and their coding agent the context of your whole codebase, across every repository your organisation has indexed, and puts changes through a plan that a person approves. Vörr (also written Vorr or Vor) is an MCP server that gives Claude, ChatGPT, Cursor or any other AI assistant that supports MCP read access to your cloud infrastructure, code, observability and CI data. CodeZero is where engineers change code; Vörr is where anyone on the team asks how the running system behaves. Both share the same code index.

Engineering teams whose systems span many repositories, services and cloud accounts. CodeZero is for developers and the engineering leads who review their changes. Vörr is for developers, SREs, platform teams and engineering leaders who need answers about production without opening every console.

No. CodeZero works with the coding agent your developers already use (Claude Code, Codex or GitHub Copilot) on that agent's own subscription or credentials, and adds the context, planning and guardrails around it. Vörr is added as a custom connector to any AI assistant that supports MCP, including Claude, ChatGPT and Cursor. Neither product asks you to switch AI provider.

For CodeZero: install the extension from the VS Code Marketplace, sign in with your Frigga account, connect GitHub, GitLab or Bitbucket and pick the repositories to index. For Vörr: sign in at ai.frigga.cloud, complete your company profile, connect source control, cloud accounts and any observability tools, then add Vörr's MCP URL to Claude, ChatGPT, Cursor or any other assistant that supports MCP. See the setup guides.

Pricing is set with each team. Book a call from the Book a Demo page or email hello@frigga.cloud and we'll walk through it. See Book a Demo.

Frigga Cloud Labs is a software R&D company founded in 2023 and based in Bengaluru, India. It builds AI-native tools for engineering teams: CodeZero and Vörr, plus the DevOps products Shankh, Juno, Drasil, Dvarpala, Runes and Brokk. See About Us.

Yes. Frigga Cloud Labs is SOC 2 and ISO 27001 certified. Reports are available under NDA on request: email hello@frigga.cloud.

Yes. Our privacy policy sets out how personal data is processed, the rights you have under GDPR and the safeguards used for international transfers, including Standard Contractual Clauses. See Privacy Policy.

Email hello@frigga.cloud or book a call from the Book a Demo page, and we'll share our security documentation and answer your security questionnaire. See Book a Demo.

It adds an AI agent and a full visual Git client to VS Code. The panels cover chat with the agent, a Git Tree (commit graph, inline blame, compare, conflict resolution, rebase, stash, squash, branches and pushing across repositories), function tracing, the code graph, files across repositories, and the tasks of the plan being worked on.

Visual Studio Code, version 1.90 or later. CodeZero is published as a VS Code extension; there is no JetBrains plugin today. See Install and connect.

Claude Code, Codex and GitHub Copilot. CodeZero uses the agent's own subscription or credentials, so usage is billed by that provider on your existing plan, and it uses whichever model your agent is configured to use.

Yes. Every repository your organisation has connected and indexed is available to search, read and trace, whether or not it is on your machine. Indexed repositories are read-only; when an approved plan needs to change one, you check it out and the agent works there. See Working across repositories.

Organisation-wide code search, file reading, symbol definitions, references, call hierarchies, a dependency graph at directory or file level, recent changes and file ownership, plus a repository overview covering structure, dependencies, environment variables and HTTP endpoints. The graph also infers which service calls which by matching request paths to routes in other repositories, labels each inferred link with a confidence level, and lets your team confirm or correct it. See Code Graph.

TypeScript, JavaScript, Python, Go, Java, Ruby and C# are fully parsed. Kotlin, PHP, Rust, Scala, Swift and other common languages are indexed with lighter, pattern-based extraction.

GitHub (through the GitHub App or OAuth), GitLab, including self-managed GitLab, and Bitbucket Cloud.

No. The organisation index holds committed code. Your uncommitted changes stay on your machine, where the agent reads them directly from your working tree.

Yes. CodeZero's organisation code tools are also served over MCP, so an MCP-capable agent in the terminal, such as Claude Code, Codex or GitHub Copilot, can search and trace the same index.

They decide how changes are made. Auto makes small, clearly named changes directly, showing each edit for approval, and sends anything larger through a plan first. Planned always plans first and blocks edits until the plan is approved. Normal makes no plan and asks you to confirm each change. Coffee Mode carries an already-approved plan through to the end without step-by-step confirmations, but never commits, pushes, switches branches, rewrites Git history, publishes releases or runs destructive operations. See Execution modes.

The agent reads the code first, then asks you the decisions that are yours to make and records your answers. It writes a plan, with the repositories it will change and the commands that will verify it, and submits it for review. Once a person approves it, the work is split into tasks, and each task is only marked done with evidence of what was run and what it returned.

No. Approval only happens through the CodeZero panel, by a person; none of the agent's tools can approve a plan. Until a plan is approved, the agent can write the plan document and nothing else, and it cannot run shell commands.

Commits the agent creates carry a "Co-authored-by: Code-Zero <codezero@frigga.cloud>" trailer, so they are visible in your Git history and on GitHub.

Every edit or command outside an approved plan is shown to the developer for approval, with the diff, before it happens. After approval, the agent can write only in the repositories the plan named and run only the verification commands the plan declared without asking. Writes outside those repositories and to sensitive paths such as ~/.ssh are always refused. Pushing and publishing releases can never be set to "allow always", and a prompt nobody answers is treated as a refusal.

Yes. Work items are shared across your organisation, showing who opened each one, who approved its plan and how far its tasks have got.

Organisation admins have a CodeZero usage page in Frigga Accounts showing time spent, tokens, cost and tool calls per user. Members don't see each other's individual usage.

Yes. Each work item keeps the questions the agent asked and the answers given, the plan and its summary, the commands it declared for verification, who approved it and when, and each task's status and completion evidence.

It depends on the indexing mode. With managed indexing, Frigga makes a shallow clone of each repository you select onto its infrastructure to build the index. A shallow clone copies only the latest commit of a single branch: it carries none of the repository's earlier commit history and none of its other branches, and updates fetch just the newest commit again. With the self-hosted indexing agent, which runs in your environment with Docker or Helm and makes outbound connections only, Frigga never clones your repositories: the agent uploads a symbol map (names, kinds, file locations and imports, no source text) and serves individual file reads on demand.

The extension signs in through your browser using OAuth with PKCE. Tokens are kept in VS Code's SecretStorage, which uses your operating system's keychain. Access tokens last about ten minutes and refresh automatically, and the account must hold a CodeZero seat.

They are encrypted at rest with AES-256-GCM and are never returned by the API. Tokens and local paths are scrubbed from error messages before an AI model sees them, and the self-hosted indexing agent keeps its Git credential in its own environment instead of writing it into repository configuration.

Every request is scoped to the organisation of the signed-in user, taken from their authenticated identity and never from anything the request or the AI model supplies. Repository access is checked against your organisation before any code is returned.

Organisation admins manage the organisation and see its usage; members and viewers use CodeZero without access to admin or per-person usage data.

No. Semantic search uses an embedding model that runs inside Frigga's own code service; no external embedding API is called.

No. The CodeZero agent loads only CodeZero's own tools, not the MCP servers configured on the developer's machine, and every tool call passes through the same plan guard and approval rules.

The code indexing can: the self-hosted indexing agent runs in your own environment with Docker or Helm. The CodeZero service itself runs as Frigga's managed cloud service.

Vörr is an MCP server. You add its MCP URL as a custom connector to Claude, ChatGPT, Cursor or any other AI assistant that supports MCP, and ask your questions there; the assistant calls Vörr's tools to fetch facts about your systems. The Vörr web app is where you connect systems and browse Cloud Mapping, scan results and the Architecture map. See Add Vörr to Claude or ChatGPT.

Questions about cloud resources and their blast radius, what is exposed to the internet, how services depend on each other, and what changed recently. It also answers questions about code across your repositories and about metrics, logs, traces, errors, alerts and deploys, and can work toward a root cause by correlating them. See Ask Vörr.

Only if that area was scanned successfully. Vörr reports scan coverage, and separates a scanned-and-empty result from a resource type it doesn't index and from a scan that was incomplete because of access, failure or throttling, instead of reporting a false zero.

No. Vörr indexes committed code in the repositories you connect. Local, uncommitted work on a developer's machine is never read.

Yes, with Shankh. When an alert fires, Vörr gathers evidence from your connected tools and an AI model explains the likely cause. The diagnosis uses an Anthropic or OpenAI key you provide. See Shankh.

AWS, Google Cloud and Azure, across hundreds of resource types. Amazon EKS clusters are also scanned inside the cluster through a read-only Kubernetes client, and Vörr can read Terraform state. Prometheus and Nagios can be added during cloud setup. See Connect cloud accounts.

37 read-only connectors. Metrics: Prometheus, VictoriaMetrics, Thanos, Grafana Mimir, Datadog, New Relic and Dynatrace. Logs: Grafana Loki, Elasticsearch, Kibana, Graylog, Splunk and OpenObserve. Tracing: Jaeger, Zipkin, Grafana Tempo, OpenTelemetry and SigNoz. Errors: Sentry and BugSnag. Dashboards, alerting and uptime: Grafana, Alertmanager, Uptime Kuma, Nagios XI, Nagios Core and Middleware. CI/CD and delivery: Jenkins, CircleCI, Buildkite, Drone, Argo CD, Flux CD and Devtron. Artifacts, quality and infrastructure as code: Harbor, Nexus, SonarQube and Terraform. See Connect observability tools.

GitHub, GitLab, including self-managed GitLab, and Bitbucket. See Connect repositories.

On AWS, Vörr assumes an IAM role you create, protected by an external ID, with a read-only discovery policy of Describe, List and Get actions; you choose which optional read-only additions to include, such as observability, backup, CI/CD and cost data. Access keys are also accepted. On Google Cloud it uses a service account, and on Azure a service principal client secret. See Connect cloud accounts.

Only in two optional, opt-in ways. If you grant the tag-write permission, available on paid plans, Vörr can apply its own fg:-prefixed tags to resources, and every tag change is previewed and needs explicit confirmation. If you enable host inspection and give consent for it, Vörr can run read-only diagnostic commands on EC2 instances through AWS Systems Manager. Without those, Vörr's access is read-only. It never pushes code or changes your repositories or CI.

No. Vörr never calls the APIs that return secret values. For Secrets Manager and SSM parameters it records metadata only, such as names, tags and rotation settings.

No. Connecting an assistant goes through an OAuth 2.1 consent step tied to your user and organisation. For each session, Vörr's tools use a short-lived, read-only token scoped to a single organisation, and your stored cloud, Git and observability credentials are never passed to the assistant. Removing the connector from the assistant revokes its access.

Cloud, connector and Git credentials are encrypted at rest with AES-256-GCM, masked in the interface once saved, and never returned in decrypted form.

With a passwordless one-time code sent to their work email. Assistants connect through OAuth 2.1 with PKCE, using signed access tokens that expire after an hour and refresh tokens that expire after 30 days.

Members can ask questions and browse the resources in the workspaces they belong to. Administrators can also connect Git providers, cloud accounts and connectors, create workspaces and environments, and edit service maps. Workspaces scope what each team sees, and MCP authorisations are tied to the user who granted them. See Roles and permissions.

Each MCP session is bound to one user and one organisation, and every query is scoped to that organisation. Connections can be further limited to selected workspaces.

Yes. Every tool call is logged with the organisation, the user and their role, the tool, the AI client, the outcome and the duration. Secrets and payloads are never written to the log.

For questions asked over MCP, none: Vörr's tools return facts and the assistant you connect does the reasoning. Two features use a model directly: company-profile discovery during onboarding uses an OpenAI model, and alert diagnosis uses the Anthropic or OpenAI key you provide.

Vörr doesn't train models: its MCP tools return facts, and the assistant you connect does the reasoning under your agreement with that provider. For Frigga's formal training-policy statement, contact hello@frigga.cloud.

Vörr runs as Frigga's managed cloud service. With managed indexing, the repositories you connect are shallow-cloned onto Frigga's infrastructure: only the latest commit of a single branch is copied, with none of the earlier commit history or other branches. Code indexing can instead run in your own environment through the self-hosted indexing agent, so your repositories aren't cloned onto Frigga's infrastructure at all.