Operator edition · Version 1.0

Deploy PTT Talk
with evidence.

A complete path from clean checkout to a private-team instance—covering infrastructure, mobile signing, encrypted voice proof, security, recovery, and a machine-executable AI-agent contract.

PTT Talk 0.2.0 (45)Protocol 1.1AGPLv3
PTT
K3s Cloudflare Android iOS
PDFComplete operator guide22-page printable runbook PPTTeam briefing deck14 editable presentation slides MDAI-agent runbookFull build, install, and test contract

Decision 01

Choose one authoritative back end.

Both paths preserve the mobile protocol and end-to-end encrypted media contract. Select one origin, one membership authority, and one recovery model for each private-team instance.

Supported self-hosted target

K3s + Helm

For teams that control their network, cluster, storage, and recovery process.

  • Rust control and UDP relay services
  • PostgreSQL, Redis, and MinIO
  • Authenticated UDP with encrypted TLS fallback
  • Helm-managed health, network policy, backup, upgrade, and rollback
Managed-edge alternative

Cloudflare

For teams that prefer managed edge services and encrypted TLS media.

  • Workers and Durable Objects
  • D1, R2, Queues, and dead-letter queues
  • Encrypted WebSocket/TLS media only
  • Independent staging and production resources
No plaintext fallback. Crypto or authorization failure stops transmission or forces an upgrade.

Prepare once

Bring every dependency before touching production.

Missing delivery, storage, signing, or recovery inputs create deployments that look healthy but cannot serve a real team.

01

Domain + TLS

Public HTTPS origin, K3s gRPC name, valid certificate SANs, TCP 443, and UDP 47000 where applicable.

02

Email + push

Verified sender, least-privilege FCM identity, and independent APNs production and sandbox keys.

03

Durable data

Live storage, an actually encrypted backup destination, retention policy, and named recovery owner.

04

Signed clients

Apple entitlements and profile, Android upload key, Firebase client configuration, and matching app-link associations.

05

Test fleet

Two Android and two iOS physical devices plus an external microphone for audible acoustic proof.

06

Authority

Named operator, exact account or cluster, controlled change window, rollback plan, and incident ownership.

Build and install

Promote one commit. Never rebuild between stages.

Validate source, produce immutable artifacts, prove them in staging, and promote the exact same bits to production.

  1. 01
    Freeze source

    Record the full Git SHA, expected remote, clean status, product version, protocol version, and toolchain.

  2. 02
    Run Promptfoo campaigns

    Promptfoo orchestrates the protocol, Rust, Kotlin, Swift, TypeScript, browser, integration, recovery, and security gates while native tools produce deterministic evidence.

  3. 03
    Build immutable artifacts

    Create multi-architecture control, relay, and admin images plus signed Android and iOS release artifacts.

  4. 04
    Render before apply

    Review DNS, TLS, public exposure, storage classes, image digests, migrations, secrets handling, and rollback.

  5. 05
    Install staging

    Use an atomic Helm release on K3s or independent Cloudflare staging resources. Wait for readiness and delivery providers.

  6. 06
    Promote proven bits

    After the full release gate passes, deploy those exact digests and hashes to production without rebuilding.

AI-AGENT ENTRY POINTdocs/DEPLOYMENT_GUIDE.md

The Markdown guide includes commands, protected-variable contracts, expected evidence, stop conditions, and a JSON handoff manifest.

Open the machine-readable guide

Campaign definitions, privacy controls, route accounting, and physical proof requirements are maintained in the Promptfoo testing contract.

Verification

Build confidence in layers.
Finish with real acoustics.

A green build is necessary. It is not proof that encrypted audio reached another device's speaker.

05 · Soak + security + restore
04 · Four physical devices + audible proof
03 · Deployed two-client encrypted media
02 · Simulators + accessibility + lifecycle
01 · Unit + wire + integration + Helm
Every direction

Android to Android, iOS to iOS, and Android to iOS in both directions.

Every state

Foreground, background, locked screen, push wake, reboot, restoration, and rapid repress.

Every network

Wi-Fi, mobile data, transitions, UDP available, and UDP blocked with encrypted TLS fallback.

Every route

Speaker, receiver, wired, supported Bluetooth, interruptions, and route changes.

MeasurePress → grant → encrypted frame → decode → audible playback Good Wi-Fi/LTE targets: warm floor p95 under 150 ms · mouth-to-ear p95 under 400 ms

Operate

Recovery is part of deployment.

The team must be able to restore, rotate, upgrade, and fail closed before a production handoff is complete.

01

Backup together

Capture the database and ciphertext objects at one recovery point. Encrypt, retain, inventory, and verify.

02

Restore in isolation

Stop writers, snapshot current state, restore both parts, then prove administration, history, push, and voice.

03

Upgrade atomically

Promote tested digests, inspect the manifest diff, apply atomically, and rerun health and communication gates.

04

Respond safely

Disable affected transmission, revoke and rotate, preserve redacted evidence, and never downgrade encryption.

Final gate

Missing evidence is a no-go.

  • Exact commit and immutable digests
  • Email, FCM, and separate APNs paths
  • Two accounts with two devices each
  • Four-device audible voice matrix
  • Chat files, voice notes, and video
  • Eight-hour Android screen-off soak
  • Coordinated backup and restore drill
  • Security review and incident ownership

Ready to plan your instance?

Use the complete runbook.
Keep the evidence.

Download the operator guide, brief the team, and give your automation agent the exact same source of truth.