- Home
- Skills
- Agents & Orchestration
- E2EE Matrix for AI Agents
Works with the AI Agents you already use
E2EE Matrix for AI Agents
Hardened deployment playbook and failure-mode catalogue for E2EE Matrix agents and encrypted voice runtimes.
$19
E2EE Matrix for AI Agents
Example session with this skill installed
My Matrix agent is answering calls but the caller hears nothing, even though LiveKit shows audio activity. How do I fix this?
- Read your context and instructions
- Compiled the e2ee matrix
The most likely cause is a participant identity mismatch or a missing key installation on the encrypted path. In the MSC4143 generation, the caller's client will not play audio unless the identity exactly matches the membership suffix.
Matrix Voice Lane Diagnostic
| Component | Check | Requirement |
|---|---|---|
| Identity Suffix | LiveKit Participant ID | Must be <mxid>:<device> (e.g. @bot:host.com:DEVICEID) |
| RTC Focus | /.well-known/matrix/client | Must return valid org.matrix.msc4143.rtc_foci JSON |
| Encryption | EncryptionType | Must be explicitly set to GCM with HKDF derivation |
| Key Transport | to-device events | Check logs for media key installed within 5s of answer |
| Receiver Binding | Room Binding | Ensure receiver is bound unconditionally on the encrypted path |
Next steps
- Verify the
livekit_aliasin your foci config is the Matrix Room ID, not the SFU room name. - Run the §3.7 negative test: withhold keys and confirm the call aborts rather than publishing plaintext.
- Check the JWT sidecar logs to ensure the minted room name matches the
sha256derivation of the room ID and slot.
Connects securely to your tools. The creator never sees your data.
About this skill
Your agent deserves a real, encrypted home on Matrix — not a plaintext webhook relay. This is an operations playbook and battle-tested version matrix for running an AI agent on a self-hosted Synapse homeserver with real end-to-end encryption and a voice lane where callers ring the agent over E2EE-encrypted calls — self-hosted end to end, no WhatsApp, no third party in the key path.
What this is — and isn't. This is the ops layer: architecture, a tested version matrix (Synapse 1.162, LiveKit 1.13, matrix-nio 0.26 + vodozemac 0.10), config-level procedures with restart gates, an acceptance checklist, and a failure-mode catalogue distilled from a production deployment. It is NOT a turnkey agent implementation and ships no runtime code — the runtime side is covered by an open-source reference implementation (bellmouth, Apache-2.0) that implements exactly the interfaces this playbook describes. Deploy the stack here, run the reference agent there, and you have a complete E2EE voice-agent setup.
The value is the failure modes. Every entry in the troubleshooting catalogue was observed in production: ghost devices that silence strict clients, one-time-key piles that wedge uploads, the device-token mismatch that silently rejects every key upload, and sync loops that freeze while every process check stays green. Each row maps a symptom to its observed root cause and the fix that resolved it — labeled as observed causes, with the discriminating check to run before any destructive recovery.
What you get
- A deployment sequence with restart gates, config-validation preflights, and rollback points — pinned to a tested component matrix
- An E2EE key setup chain that distinguishes discovery, signature validation, and trust bootstrap — fail-closed by default, with explicit voice-media encryption parameters (GCM + HKDF + pinned salt) and a mandatory negative test proving plaintext publication is impossible
- A voice-lane recipe: MatrixRTC calls, VAD/STT/TTS hardening, barge-in
- A failure-mode table: 31 symptoms mapped to observed root causes and fixes
- A recovery ladder where no step is "harmless" — restore-first discipline, with a single shared device-deletion gate
- An operational acceptance checklist (rollout gates) you can run before declaring the voice lane production-ready
Who this is for
Self-hosters and AI-agent builders who want their agent living on Matrix with real E2EE and working voice calls — inheriting a proven ops runbook instead of rediscovering every silent failure mode themselves, with an open-source reference implementation to build on.
MIT licensed. Tested against the exact version matrix inside (re-verify version-sensitive steps on upgrades — flagged inline with ⚠).
Tested for Synapse 1.162.0 and matrix-nio 0.26.0. No server-side key backup download. Requires manual derivation for MSC4195-compatible clients.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 13 days ago
Requires