The trust model

Available today

Meetless Agent captures a lot and trusts almost none of it by default. That is the point. The rule is simple:

Ingest generously. Promote conservatively.

Why this matters

An AI coding agent produces claims constantly: inferences, proposed decisions, draft specs, intermediate plans. Some are right, some are stale, some are wrong. If you let everything the agent says become truth, you poison the well. The next session then grounds itself on the last session’s guesses, and the errors compound.

So Meetless Agent separates capture from trust.

Evidence, not truth

Everything a session produces is captured as evidence: prompts, decisions, tool calls, and any documents the agent writes. Evidence is indexed so it is searchable, but evidence is not trusted. It is born pending.

Note

Pending knowledge never grounds an answer or an injection. It sits in a review queue until a human decides. In mla terms, an ingest is born with review outcome PENDING and becomes usable only after it is accepted.

Trust comes from the path, not the label

You do not get to trust a claim by labeling it trustworthy. Meetless Agent derives trust from how the knowledge arrived, not from a flag the writer sets. An agent cannot mark its own output “trusted” and have that stick. Every ingest is born pending regardless of what it claims about itself, and only human promotion moves it. This is deliberate: it removes the obvious attack where the thing being governed edits its own trust level.

Sources are not equal

Not all evidence carries the same weight, and Meetless Agent does not pretend it does:

• Existing accepted knowledge is the highest authority. It already passed human review.
• Live user intent in the current session is high authority for that session.
• Code facts are strong for what exists, weaker for why it exists.
• Agent claims are low authority until backed by code, an artifact, or a human.
• Produced artifacts are evidence, not canon. A generated spec is a draft until someone accepts it.

The serving law

Here is the guarantee that makes the output trustworthy:

Only human-approved knowledge grounds injection and answers.

When you accept a piece of knowledge, it starts shaping what the agent sees and what a query returns. Until then, it does not. Nothing the agent asserted about itself becomes truth without a human gate.

This is why you can trust a Meetless Agent answer: every claim it serves traces to a source a human approved. The mechanics of accepting and promoting are covered in Reviewing & promoting knowledge.