Use cases Memory Modes
How teams build trust

One memory platform.
Multiple trust models.

Every team has a different comfort level with AI memory.
Tenure meets you where you are, and evolves as you're ready.

Document-Driven Docs only, no learning
Curated AI proposes, humans approve
Adaptive Evolves with the team
Reflective Insights, no injection

These aren't tiers. They're different philosophies. You can mix them, run them in parallel, or start with one and expand as your team gains confidence.

01 Document-Driven Memory
Enterprises Regulated environments Teams starting with AI memory

Extract beliefs from existing documentation.
Inject only approved knowledge.

For organizations that want AI grounded in Confluence, ADRs, READMEs, and coding standards. Nothing the model receives was learned from conversation or behavior. It was written down by a human and explicitly approved.

Sources
  • Confluence
  • ADRs
  • Internal docs
  • Repository documentation
Memory Behavior
  • No conversational learning
  • No automatic updates
  • Only documented knowledge reaches the model
The model knows exactly what your documentation says. Nothing it hasn't been given.
02 Curated Memory
Growing teams Engineering organizations Teams that want oversight

AI proposes beliefs.
Humans decide what becomes institutional knowledge.

Tenure continuously discovers conventions and patterns from documentation, code, and conversations. But nothing reaches the model until a human approves it. The memory evolves intentionally, not automatically.

Sources
  • Documentation
  • Code patterns
  • Conversations
Memory Behavior
  • Suggested beliefs require approval
  • Full evidence and confidence shown
  • Institutional memory evolves intentionally
Proposed belief
Use Zod for validation.
· 9 PR reviews · 84 code generations · 2 team discussions
0.88
03 Adaptive Memory
Fast-moving teams Startups AI-first organizations

Continuously learn from how
the team actually works.

Tenure discovers conventions, decisions, and expertise automatically. Context is retrieved per turn, so topic switches don't drag irrelevant memory into the conversation. Memory doesn't sit static. It reflects how the team actually builds.

Sources
  • Documentation
  • Conversations
  • Generated code
  • Team interactions
Memory Behavior
  • Beliefs evolve naturally
  • Retrieval happens per turn
  • Context changes with the task
How memory shifts over time
A recurring error pattern becomes the dominant belief about error handling.
gRPC gradually replaces GraphQL as the team's preferred transport.
New expertise around event sourcing surfaces as a team-level belief.
04 Reflective Memory
Engineering leadership Knowledge management Organizations exploring AI memory

Generate insights without
injecting memory into models.

Turn work into weekly summaries, expertise maps, and organizational insights. The model never receives memory. You do. Reflective Intelligence is for teams that want to understand what's happening before deciding whether to act on it.

Sources
  • Conversations
  • Code generation
  • Team interactions
Memory Behavior
  • No memory injection required
  • No behavioral changes to AI
  • Human-facing summaries and trends
Insights include
Emerging expertise
~ Architectural drift
! Repeated friction points
? Knowledge gaps
Mix and match

Learning and injection are independent.

You don't have to learn from everything you inject from, or inject everything you learn from. Configure each axis separately, per team or per project.

Learn From
Docs
Code
Conversations
Inject From
Docs
Approved beliefs
Conversations

Start conservatively. Expand as trust is earned. The configuration is yours to change at any point.

The real question

How much control do you have?

You set the terms. You decide what gets learned, what gets injected, and what stays human-facing.

Some teams begin with documentation.
Others want human approval on every belief.
Some embrace continuous learning.
And some use Tenure purely for insight.

Trust isn't binary. It's configurable. Tenure is built for teams that want to set the terms.