Provable oversight
for your AI agents.

meander reads what your agents do and decides what may pass. Every decision comes from rules a human approved, and every verdict carries its evidence.

100%
of verdicts carry their proof
Plan allowed? Proven. Based on Rule R-2, from 3 facts with sources.
meander · live triage
Records
Record 0041proven
Record 0040needs review
Record 0039proven
Record 0038advisory
Record 0037proven

Record 0041

Reported by agent · 2 minutes ago
Conclusion: allowedBased on Rule R-2
statusconfirmedsource: tool call
assigned_toreviewer-7source: operator
checkedtruederived · Rule R-2
meander proposes the world model
Recordentity · 3 fields
statustext · from telemetry
assigned_torelation · from telemetry
checkedflag · derived by rules
Approve Reject

Connect an agent. meander learns the world.

Point your agent's telemetry at meander. It reads the structure of what the agent touches, proposes the world model, and you approve it before anything counts.

Everything oversight needs, in one place.
Connect, model, decide, and prove.

Connect

Agents stream their telemetry to meander. Read-only: meander never acts on your behalf.

World

Facts with provenance: every value knows which tool call or person it came from.

Rules

Policies you author and approve decide what is allowed. No rule ships without a human.

Proof

Each verdict links to the rules and facts behind it, down to the audit trail.

Connect your agents, model the world, decide by rules, and prove every verdict.

Connect
Connect an agent
Endpointhttps://…/v1/traces
Source keyshown once · sk-…
SDKpip install meander-agent
Statusreceiving spans

Connect the agents you already run.

One endpoint and a source key. Your agent keeps working as before; meander reads its telemetry and never calls anything itself.

World

A world model built from evidence.

Every fact carries where it came from: which tool call, which person, when. Claims by an agent never count as facts on their own.

Record 0041 · facts
statussource: tool call
assigned_tosource: operator
claimed by agentnot authoritative
checkedderived · Rule R-2
Rules
When a record's status is confirmed
and it has an assigned_to
then conclude checked
active  approved by you · applies to all records

Decisions come from rules you approved.

Author rules in plain language, preview their effect in a sandbox, then activate. Agents can propose rules; only people can make them law.

Proof

Every verdict carries its proof.

Ask why, and meander shows the exact rules and facts behind a conclusion. The audit log keeps every decision, forever.

Conclusion · allowed proven
Rule R-2 · approved by operator
status = confirmed · source: tool call
assigned_to = reviewer-7 · source: operator

Put your agents under provable oversight.

Access is by invitation. Leave your email and we'll send you an invite link.

Already invited? Sign in