For teams

Your team's knowledge should outlive every session.

Your team has spent years making decisions the AI has never been told about. Architecture rules. Team conventions. Lessons learned the hard way.

Tenure puts that knowledge in the room before the first message is sent.

The problem

Onboarding has two victims.

Everyone focuses on the new hire. But there's a second person paying the cost: the senior engineer who got pulled out of a focused work block to answer questions that already have answers somewhere. That context exists. It just isn't accessible to the AI, so it gets re-explained by a human instead.

Without shared memory
Why does the API use POST for search instead of GET?
That's unusual. Typically GET is used for read operations. You may want to refactor this to follow REST conventions.
No, we decided this intentionally because our search queries include sensitive filter data that shouldn't be logged in proxies.
Understood. I'll keep that in mind for this session.
This conversation happens again. Next session. Next engineer. Every time.
With shared memory
Why does the API use POST for search instead of GET?
POST is used for all search endpoints by design. Search queries contain sensitive filter parameters that should not appear in proxy or CDN logs. This is an intentional deviation from REST conventions. Do not suggest changing it.
Org belief surfaced automatically. No one had to explain it.
How it works

Three layers. Every belief knows its scope.

Not all knowledge belongs to everyone. Tenure's visibility model lets you be precise about what travels with the org, what stays within a team, and what is personal to an engineer.

org

Organization-wide standards

The non-negotiables. Architecture decisions, security constraints, naming conventions, compliance rules, and anything else that applies to every engineer on every project. Set once by a principal or lead. Surfaces automatically for everyone.

"All service-to-service calls must use mTLS. No exceptions."
"We use snake_case for database columns and camelCase for API responses."
"The payments service owns its own data. No other service queries its tables directly."
team

Team conventions

How your squad actually works. Sprint rituals, patterns your team has adopted, known quirks in the codebase, decisions made in the last planning session. Visible to team members. Doesn't bleed into other squads.

"We wrap all third-party API calls in a retry helper from /lib/resilience. Do not call external services directly."
"The checkout flow has a known race condition in the inventory lock. Tracked in PLAT-1847. Don't try to fix it without reading that ticket first."
"PRs need a test that covers the unhappy path. If there isn't one, the review will ask for it."
user

Individual expertise and preferences

Personal to each engineer. Domain expertise, preferred patterns, context about work in progress, and how they like the AI to respond. Travels with them across sessions and clients. Not visible to their team or org.

"I'm the original author of the billing module. I know where all the bodies are buried."
"I prefer exhaustive TypeScript types over implicit inference. Always make types explicit."
"I'm currently refactoring the notification service. Don't suggest changes that touch that area."

Beliefs at every layer are resolved at query time. When an engineer asks a question, the relevant org, team, and user beliefs all surface together, in the right priority order, without manual context-pasting.

In practice

The AI writes code the way your team writes code.

An engineer opens VSCode. Tenure is already connected as their OpenAI-compatible provider. Before the first token is generated, the agent has the org standards, the team conventions, and that engineer's personal context in scope. What comes out looks like it was written by someone who has been on the team for months.

PRs that don't need teaching moments

The generated code already follows your patterns. The reviewer isn't correcting style or re-explaining architecture. They're reviewing logic, which is the point.

🧭

New hires that ship in week one

The new engineer doesn't need a full architectural tour before they can be useful. The AI already has the tour. They can start contributing while they're still getting oriented.

🔇

Fewer interruptions for senior engineers

The questions that get asked are the hard ones. Not "where does auth live" or "do we use tabs or spaces." That knowledge is already in the room.

📋

A living architecture record

Decisions get captured as beliefs when they're made, not documented six months later or not at all. The AI enforces them automatically until someone deliberately changes them.

Getting started

Beliefs come from real conversations.

You don't sit down and write a belief library from scratch. Beliefs accumulate from actual work. An engineer explains something to the AI, Tenure captures it. A lead adds an org-level rule in the dashboard. A team decision from today's planning session gets committed before the meeting ends.

The starting point for most teams is a short session where a principal or tech lead walks through the things they're tired of explaining. That usually takes an hour. It pays back immediately.

Deployment guide →
Capture during a real session
We use Zod for all input validation at the API boundary. Never trust raw request bodies anywhere else in the stack.
Got it. I'll use Zod schemas at API boundaries and treat all downstream inputs as already validated.
+ Save as team belief saved
And we never throw raw Zod errors to the client. Always map them to our ApiError format.
Understood. Zod errors get mapped to ApiError before any response is sent.
+ Save as team belief saved
Next engineer to ask about validation gets both of these automatically.
Worth being clear about

This is not a documentation tool.

Tenure doesn't replace your wiki, your ADRs, or your onboarding runbook. Those are reference artifacts people read deliberately. Beliefs are context the AI uses without being asked. The job is different.

A wiki page describing your auth architecture is something a new hire reads once. A belief that says "the session token is a signed JWT, validated by the gateway before the request reaches any service" is something the AI knows every time it touches anything near authentication. One is a document. The other is working memory.

Both are useful. They're not the same thing.

Get started

Your team's knowledge should travel with every session.

Deploy the Helm chart in your dev environment. Connect it to VSCode or any OpenAI-compatible client. Spend an hour capturing what your senior engineers are tired of explaining. The rest happens on its own.

MIT licensed · No call-home telemetry · Works with VSCode, Cursor, OpenWebUI, and any OpenAI-compatible client