The words this documentation leans on, defined once so the rest of it can be short.
AI dev
One concurrent unit of agent work. Buying a seat buys one AI dev, which is one runner able to execute an agent session at a time. Not a persona and not a licence for a person — a concurrency slot. See billing.
Audit log
The append-only record of everything the agent did. Each entry carries the actor, the repository, the kind of action, the input, the output and a timestamp, and is written before the action executes. Nothing updates or deletes a row. If it is not in the log, it did not happen.
Auto-merge
Merging a pull request once the machine gates agree — CI green, review passed, branch protections satisfied. It is the default on paid tiers and opt-in on free and OSS-verified accounts. Never a human rubber-stamp; either the gates pass or the PR waits.
Box
A virtual machine you own that has been enrolled into your fleet. Boxes are fungible: any box with the right capabilities can take the next task. See bring your own boxes.
BYOK
Bring Your Own Key. Inference runs against your own Anthropic, OpenAI or Gemini account. We never resell tokens and never appear in your provider invoice. See BYOK.
Capability
A token a box advertises that says what it can run — Docker, a container isolation mode, CI backfill, a language toolchain. A task declares what it requires and is only assigned to a box that satisfies all of it.
Disclosure
The statement, on every comment the agent posts, that it is an automated agent. There is no setting that removes it. Non-disclosure would be a bug.
Disposition
The verdict the reviewer reached on a finding — what it decided and why. The review page shows how dispositions line up against another reviewer's, which is what makes a migration a comparison rather than a leap.
Enrolment token
A one-time secret that lets a specific box join your fleet. It is spent on first use. In the SSH transport the operator never sees one at all. See bring your own boxes.
Entitlement
What your current seats allow: concurrent AI devs, enrollable servers, queue
depth, and the review mode. Read it from GET /v1/billing/entitlements.
Escalation
The agent stopping and notifying a human instead of acting. Security reports, licence questions, hostile threads and human-merge paths all escalate. Escalating is the safe default, not the failure case. See escalations.
Handoff
Passing a qualified issue to a coding agent — Copilot, Claude Code, OpenHands or your own — to write the actual patch. developerz.ai is a thin orchestrator: it decides and coordinates; it does not write your code.
Lane
A distinct kind of work the platform runs: triage, review, CI, releases. Lanes share the same App, the same audit log and the same policy file.
Maintainer agent
What this product is, as opposed to a coding agent. It triages, talks to reporters, hands coding off, watches pull requests and ships releases.
.maintainer.yml
The policy file committed to your repository. It bounds what the agent may do: what qualifies, what escalates, what merges automatically, what always waits for a person. If it does not parse, the agent stops rather than falling back to defaults. See the reference.
MCP
Model Context Protocol. The structured interface at mcp.developerz.ai covering
the same surface as the REST API, for clients like Claude Code. See
MCP verbs.
OSS verified
An operator-approved status for public repositories that waives billing. Applying does not grant it — a human reviews each application.
Runner
The service on a box that receives assignments and executes them. It dials out only, needs no inbound ports, and updates itself only from signature-verified artefacts.
Seat
A unit of purchase. One seat = one AI dev = one concurrency slot, and allows three enrolled servers.
Spend cap
A monthly USD ceiling on your own inference exposure. You are warned at 50% and 80%; at 100% dispatch pauses. See billing.
Thin orchestrator
The design constraint the whole product follows: coding happens in your agent, inference happens on your key, execution happens on your hardware. What we own is the decision-making and the record of it.