HAIO · Human + AI Orchestration
Agent authority
What an AI host can do, what it cannot do, and how you stay in control.
What an AI host can do, what it cannot do, and how you stay in control.
What Companion can do
✓ Read your data — with your permission
Companion can read from connected tools in read-only mode by default. It never reads more than you explicitly connect.
Example: Read Stripe to check MRR. Read Gmail subject lines to find invoices. Read Linear to understand your backlog.
✓ Propose actions — not execute them
Companion drafts Action Paths and shows you every step before anything happens. You decide what runs.
Example: Companion drafts an investor update but waits for your approval before saving it anywhere.
✓ Request write access — only when you grant it
Some Action Paths need write access (e.g. creating a Linear issue). Companion explicitly requests this. You can revoke at any time.
Example: Companion creates a Linear issue from customer feedback — only after you review the issue body and click Approve.
What Companion will never do
✗ Send messages without approval
Companion never sends an email, tweet, or Slack message without explicit human approval for that specific message.
✗ Make payments
Companion cannot authorize transactions, payments, or wire transfers. Payment approval always requires a human.
✗ Delete data
Companion never deletes files, emails, or records. Destructive operations are prohibited by default.
✗ Act on other platforms without disclosure
When Companion engages on social platforms, it always identifies itself as HUMΛN Companion — an AI agent. We don't pretend to be a human.
The Proof Card promise
Every time Companion completes an Action Path, it generates a cryptographically signed Proof Card — a permanent, shareable record of what was done, when, and with whose approval. You own this record. We cannot delete it.
See Trust methodology for how evidence and cards are published.
Related
Build with HUMΛN
Builders