Install with your coding agent
Xovris does the install for you without any agent: connect the GitHub App and approve its pull request.
1. Add the server
Section titled “1. Add the server”The first time, it registers and signs you in through your browser.
2. Paste this prompt
Section titled “2. Paste this prompt”Add Xovris monitoring to this project.1. Call the xovris tool install_plan and show me the changes it lists before making any.2. Run `npx xovris@latest init --yes` in the project folder. When it prints a sign-in link and code, ask me to approve it in my browser - never type a key yourself.3. Once I have deployed, call run_verify with the connection id the wizard printed and tell me only what it reports.If I ask to undo it, run `npx xovris@latest remove`.The agent shows each change before it makes it and never handles a key: the wizard writes it into your git-ignored env file.
What it can do
Section titled “What it can do”The server gives your agent 34 tools: twenty read and fourteen act. Each act runs under the workspace's own policy, as the person the token belongs to, and every call is recorded in the audit log.
Tools that read
Section titled “Tools that read”| Tool | What it does |
|---|---|
setup_status | The setup steps and where each connection stands, as stored |
install_plan | The exact changes the wizard will make for a stack |
run_verify | Reads a connection back – telemetry is reported only once an event is stored |
connection_list | Every connection the token can see, with its health |
incident_list | Every incident the token can see, open ones first, each with its severity, stage and when it opened - a page at a time |
incident_get | One incident with its timeline, the cause, the evidence, the fix and its proof |
incident_cause | The cause Xovris already found for one incident - its causal path, the check that decided it and what is still open, so the agent starts from Xovris’s reading, never from nothing |
why_failing | Xovris’s reading of a failing system - its cause and how sure Xovris is, what saw it and what is still open - for the incident named, else the pinned or newest open one |
incident_pinned | The incident the person pinned - the one they are working on now - in full, then every other incident they pinned; start here when asked to debug "the incident" |
issue_list | Error issues with their counts, releases and status, newest first |
review_change | Reviews a diff the way Xovris reviews a pull request: each finding with its severity, why it matters and the exact fix |
hunt_code | Reads every file handed over and raises each real defect with its fix |
review_preferences | Every review preference the workspace keeps, in the person’s own words |
code_task | Where a coding task stands: being carried out, proven with its pull request, or what was tried and what happens next |
plan_issue | Turns an issue or idea into a phased plan grounded in the code and the workspace’s live systems |
judge_plan | Checks a finished change against its plan: what it left out, what it touched that the plan never named |
ask_change | Answers a question about a change from the change and the files handed over, citing the lines it read |
connected_systems | Every system connected by its address: what it is, how many things Xovris reads on it, what it runs, and whether it answered the last reading |
ai_choices | Every model the workspace can pick, each with what a million tokens draw from its allowance or credits – the same list on every plan – and where a retired one moves |
upload_status | How many bytes of an upload Xovris holds, where a stopped upload carries on from, and whether it is stored |
Tools that act
Section titled “Tools that act”| Tool | What it does |
|---|---|
incident_pin | Pins or unpins an incident for the person, so it is the one incident_pinned reads and the one shown first on their web, phone and desktop - a change to what they see first, never to their systems |
state_review_preference | Keeps a review preference the team states, for the workspace or one repository |
edit_review_preference | Says a kept preference again in new words |
forget_review_preference | Deletes a kept review preference |
import_review_guidelines | Keeps each preference a guidelines file states, line by line |
start_code_task | A change to make in a connected repository – proven by its tests and opened as a pull request |
code_job | A coding job, in a connected repository, any public GitHub repository or a new project – worked out and proven by a test that fails before the change and passes after it |
build_project | Puts a build that code_job or xovris_do made live in one step – on Xovris, or on Vercel, Netlify or Cloudflare Pages through your own connection, said live only once its address answers – or opens it as a pull request in a connected repository, or restores it |
build_app | Everything after a build is live: Xovris stops a running build, shares it by a sealed read-only link for the hours the agent names, reads its back end – its data, sign-in, storage, payments, AI and automations – and sets or removes its secrets |
xovris_do | Takes any request and does it on the workspace’s own systems – an incident fixed, an address blocked, a check added, a service restarted, a certificate renewed, the last change undone |
connect_system | Connects any system by its address – a NAS, a camera recorder, a printer, a router, an app, or any API that publishes a description (OpenAPI, GraphQL, OData, OpenRPC, SOAP) or speaks IPP – and Xovris reads its own description and its inventory within the minute and watches each thing on it |
upload_open | Opens an upload of a file of any size – a picture, a recording, a log archive, a database dump – and answers where to send its bytes, a target that carries on from the last byte it holds if the connection drops |
upload_complete | Finishes an upload once every byte is sent – it counts only once the whole file is stored, and Xovris says to send the rest if any is missing |
repair_decide | Records the decision on a fix that waits for one, following the workspace’s approval settings |
Any agent client signs itself in: it registers with Xovris by itself (dynamic client registration, or its own client ID metadata document), opens Xovris in your browser for you to allow it into your workspace, and gets a token that reaches exactly the scopes you allowed.
Access can be limited to one project: such a token sees nothing outside it. Revoking the token in Connections stops it at once, and every call is recorded in the audit log.
Agent skill
Section titled “Agent skill”Agents that load skills can use the install-xovris skill. It follows the same steps.
Your data is never used to train our AI. How your data is kept
A problem is called fixed only once it works again. How each fix is checked
XOVRIS LTD, registered in England and Wales, company number 17499910. Registered office: 66 Paul Street, London EC2A 4NA.