MoQ News #1

We have been writing about draft-ietf-moq-transport for months — the evolution from draft-00, draft-18, draft-19, and the internals of subscriptions, delivery order, and relays.

That focus is now out of date. The transport draft is one document out of more than forty active Internet-Drafts carrying “moq” in their name, and the interesting work has moved to the layers around it: what goes inside the objects, who is allowed to subscribe, how relays are operated and observed, and — increasingly — what non-media applications want from a publish/subscribe transport.

So this is the first issue of MoQ News: a running roundup of drafts, mailing list arguments, and working group decisions, published every two weeks. This opening issue does double duty as a map, because nobody should have to read forty-four abstracts to work out which five matter to them.

If you want the blow-by-blow from Vienna, that is in our IETF 126 recap. This issue is about the document pile.

The state of the pile

The working group has seven adopted drafts:

Draft	Title	Latest
moq-transport	Media over QUIC Transport	-19, 6 Jul 2026
moq-msf	MOQT Streaming Format	-01, 2 Jun 2026
moq-cmsf	CMAF-compliant MSF implementation	-01, 3 Jun 2026
moq-loc	Low Overhead Media Container	-04, 20 Jul 2026
moq-c4m	Authorization using Common Access Tokens	-01, 18 Jun 2026
moq-secure-objects	End-to-End Secure Objects	-01, 6 Jul 2026
moq-privacy-pass-auth	Privacy Pass Authentication for MoQ	-03, 6 Jul 2026

Around them sit roughly thirty-seven individual drafts. Six of those seven adopted documents are blocked behind the transport draft’s registries — as Alan Frindell put it in Vienna, if transport does not ship, nothing else can progress.

Published since IETF 126

Three weeks, seven documents.

draft-lcurley-moq-hang-02, moq-timestamp-01, moq-cluster-00 (3 August). Luke Curley pushed three at once. Cluster is new: an extension for relay clustering, joining moq-lite, hang, largest-group, and probe in a parallel stack that deliberately trades the transport draft’s generality for a smaller surface. Whatever you think of the split, this is the most active alternative design in the ecosystem and it is not slowing down.

draft-liu-moq-feedback-00 (31 July). Yanmei Liu, Minghui Jiang, and Ronghua Wu (Alibaba / Ant Group) turn their IETF 126 presentation into a specification: receivers report per-object delivery status back to senders, reusing the existing Track/Object data model rather than adding control messages. The framing is a three-layer split — application, MoQ, transport — where QUIC’s packet-level feedback is supplemented by per-object media-semantic feedback so senders can make informed bitrate, frame rate, and pacing decisions. Relay-based topologies are the motivating case, since each hop can degrade quality invisibly. This is the most-discussed draft on the list right now.

draft-pardue-moq-qlog-moq-events -07 (late July). Lucas Pardue (Cloudflare), Mathis Engelbart (TU Munich), and Aman Sharma (Meta) define a qlog event schema for MoQT — structured logging of protocol interactions in the same tradition as QUIC and HTTP/3 qlog. Still an individual draft with no working group standing, but at revision 07 it is one of the most mature unadopted documents in the space, and the newly agreed management/metrics/diagnostics document is an obvious home for this kind of work.

We have a professional interest here: moqtap’s .moqtrace format solves the same problem with different tradeoffs (CBOR, streamable, detail-levelled). We will write up the comparison properly in a future post — the short version is that they are complementary, and a tool that reads both is strictly more useful than one that reads either.

draft-altanai-moq-relay-geocode-01 (24 July). Geographic location for MoQ relays — relay selection by locality, which is a CDN problem MoQT has so far left to deployment.

draft-ietf-moq-loc-04 (20 July). The revision presented in Vienna: public properties in headers, private properties in payloads, and a new audio config property (0x0f) carrying WebCodecs decoder configurations. Two open problems from the session: a codepoint registry collision with Secure Objects, and no property for frame duration, which breaks variable-frame-rate video.

The pile, by layer

Here is the map. Six clusters, and most readers only need one or two.

1. Transport core and extensions

The base spec plus the things that want to become part of it or hang off it.

moq-transport-19 — the base. See our draft-19 walkthrough.
draft-duke-moq-subscribe-rewind-02 — the Rewind Subscription Filter. Another entry in the filter convergence.
draft-frindell-moq-moqpack-00 — QPACK-style header compression for MoQ.
draft-nandakumar-moq-qmux-moqt-00 — MOQT over QMux.
draft-lcurley-moq-largest-group-00, moq-probe-00, moq-timestamp-01, moq-cluster-00 — small, focused extensions.
draft-westerlund-moq-overview-00 — an architectural overview of the whole system. If you are new to MoQ and want one document rather than forty-four, start here.

Not yet a draft but coming: the management, metrics, and diagnostics document the working group agreed in Vienna to spin out of transport, with Alperen Temel, Alan Frindell, and Ian Swett attached.

2. What goes inside the objects

MoQT deliberately says almost nothing about payloads. This cluster fills that in, and it is the fastest-growing part of the ecosystem.

MSF — the MOQT Streaming Format, the base layer for media on MoQT.
CMSF — CMAF-compliant MSF, for anyone with an existing CMAF pipeline.
LOC — Low Overhead Media Container, the minimal-overhead alternative.
LOCMAF -01 — Low Overhead CMAF, splitting the difference.
Packaging drafts for specific formats: WebVTT and IMSC1 subtitles, MPEG-2 Transport Stream, and NMSF for neural video codecs.
SCTE-35 over the MSF Event Timeline — ad insertion signalling. If you work in broadcast, this is the draft that decides whether MoQT can carry your existing ad workflow.

The live argument in this cluster, from the Vienna MSF session, is about codepoints: standardized elements have been squatting in the application-specific (private) range, and Will Law and Mo Zanaty took the action to move track and object configuration properties to standardized codepoints. Anyone who has already shipped against the private range should watch that work.

3. Authorization and security

Zero coverage on this site so far, which we intend to fix.

C4M — authorization using Common Access Tokens.
Privacy Pass Authentication -03 — unlinkable authorization tokens.
Secure Objects -01 — end-to-end object encryption. Proposed for working group last call in Vienna, then parked pending the transport registry. Likely to be the first MoQ draft to reach last call.
draft-nandakumar-moq-generic-dpop-proof-00 — an application-agnostic DPoP framework, relevant because the Auth Design Team is leaning toward a DPoP-style challenge-response.

The Auth Design Team reports at an October interim. The open question is where the challenge-response is signalled — inline on the subscription, or on a dedicated session-level stream — and the design constraint to remember is Ted Hardie’s: relays that must maintain subscriber databases are a privacy liability.

4. Relay operations

The cluster that matters if you run MoQ rather than implement it.

draft-englishm-moq-relay-dos-01 — denial-of-service considerations for relay deployments.
draft-englishm-moq-cdn-provisioning-00 — CDN provisioning.
draft-altanai-moq-relay-geocode-01 — geographic relay location.
draft-evens-moq-bench — a benchmarking methodology for MoQT relays under realistic conditions.
draft-defoy-moq-relay-network-handling-05 — relays for high-throughput low-latency traffic in 5G networks.
moq-qlog event definitions -07 — structured logging.

Read alongside the IETF 126 interop numbers, this cluster is the ecosystem admitting that relay behavior under load and under attack is underspecified. The concurrent-subscriber failures in the interop report are exactly the kind of thing a benchmarking methodology exists to catch.

5. Conferencing and synchronized playout

Applications built directly on MoQ, presented in Vienna and now filed as drafts.

MOCHA — Cullen Jennings’ six-draft suite for decentralized conferencing without SFUs: Chat, Meetings, Reactions, Identity, MLS Keying, and a Personal Address Book. All published 6 July.
TEMPO — Suhas Nandakumar’s timing extension for synchronized playout across clients, using capture timestamps, playout delay, and relay-updated hop times.
draft-nandakumar-atproto-atom-00 — the AT Protocol over MoQ.

MOCHA is worth understanding even if you never build conferencing, because it is the strongest argument in the Top N Tracks debate: if relays can do active-speaker selection as a filter, multi-party conferencing needs no media server at all.

6. AI agents

The newest cluster, and the one most likely to bring people to MoQ who have never streamed video.

At IETF 126 the agentproto BoF met as a WG-forming session on 23 July, building on work by Jonathan Rosenberg and Cullen Jennings. Its layered architecture places an agent session layer between applications (MCP, A2A) and transport — and names MoQ, alongside WebTransport, QUIC, and WebRTC, as a transport option. The stated problem: HTTP is stateless, WebSocket has no multiplexing, priority, or partial teardown, and MoQ is promising but not agent-aware.

The drafts:

PACE — Protocol for Agent Communication Exchange.
MCP and Agent Skills over MoQT — Model Context Protocol carried over MoQ.
draft-liu-moq-live-agent-interaction-01 and draft-liu-agent-protocol-over-moq-00.
draft-nandakumar-ai-agent-moq-transport-00 — MOQ Transport for Agent Protocols.

Treat all of this as pre-standard. A BoF is not a working group, and none of these drafts is adopted. But the direction is unmistakable, and it is already pulling on the transport spec: the Feedback draft names agent interaction as a motivating use case alongside live media, and the IETF 126 performance presentation was titled “MoQ Performance and Feedback Requirements for Live Media and Agent Interaction.”

The 10 August interim

The working group met virtually on 10 August (interim-2026-moq-21): FETCH pacing, chair slides, and a working session on open PRs and issues. Minutes were not posted as of writing, so what follows is the published agenda rather than the outcomes.

The most useful thing in it is an explicitly ordered queue of in-flight PRs for draft-20:

The agenda notes that items 3 and 4 need updating once the first two merge. That ordering is the clearest public statement yet of what gates draft-20: the filter rewrite is the long pole, and Fill Fetch and SWITCH_FROM are downstream of it.

Worth knowing where #1809 actually stands, because it is easy to assume a much-discussed PR has landed. It has not. It is still open, carrying approvals from Ian Swett, Alan Frindell, and Suhas Nandakumar, with unresolved change requests from Cullen Jennings. Anyone writing filter code against the current enum is still writing against a design the working group intends to delete.

Two more PRs were up for review: PUBLISH contains subscription params #1834, reflecting the Vienna discussions, and Publisher default priority update #1770 — a proposal originally requested at the June London interim, authored by Michal Hosna. The latter makes DEFAULT_PUBLISHER_PRIORITY an updatable hop-by-hop parameter rather than a fixed initial value; the review discussion accepts some racy edge cases, resolves conflicting priorities for the same object in favour of the lower value, and leaves open whether relays should reprioritize matching subgroups when the default changes mid-stream. If you have read our delivery order and priority post, this is the loose end it did not cover.

Three issues were tabled for discussion — #1453 Send Rate parameter (Will Law presenting a brief proposal), #1352 whether SUBSCRIBE still needs a forward parameter once filters exist (Suhas wrote a PR to demonstrate feasibility), and #1801 reconsidering “OR” functionality in range filters, filed by Victor against draft-19. All three are filter-adjacent, which tells you where the working group’s attention is.

Three new issues were filed: #1842 (subgroup timeout overrides not propagated by relays joining mid-process), #1840 (INCLUDE_PROPERTIES=0 interacting with mandatory track properties), and #1835 (is the query component of a URI scoping?).

On the mailing list

The moq list has been busy. Threads worth reading:

“Pacing for FETCH” (Will Law, 4 August). Follow-up to the Fetch Pacing presentation in Vienna, and presented again at the 10 August interim. FETCH can dump a large backlog at a subscriber as fast as congestion control allows, which is rarely what anyone wants when the subscriber is also receiving live content on the same connection.

“New Version Published: Feedback draft” (from 3 August). The most active thread on the list, with Cullen Fluffy Jennings, Ronghua Wu, Suhas Nandakumar, and Yu You (Nokia) all responding within two days of publication.

“Top Tracks and SSTS (or ABR in general)” (Ian Swett, 27 July, with replies from Yu You and Will Law). The substantive question behind the procedural fight: whether Top N Tracks and Sender-Side Track Switching are two mechanisms or one, and what MoQT’s adaptive bitrate story actually is. Mo Zanaty’s “Top Tracks Filter moved to PR#1830” (24 July) is the tracking pointer.

“MoQT relay diagnostics follow-up” (Altanai B, 23 July) — feeding the new diagnostics document.

Also worth subscribing to: the Weekly GitHub digest bot, which summarizes repository activity across the working group’s documents and is the cheapest way to keep up without reading every PR.

Dates ahead

When	What
24 August	Virtual interim (agenda still TBD as of writing)
Early September	Virtual interop day, targeting draft-18
8 and 21 September	Further virtual interims on the calendar
14–15 October, Seattle	In-person interim — Auth Design Team on challenge-response signalling; target: implementations on draft-20 (superseded on 10 August — see the update below)
Unscheduled	Virtual interim on SSTS packaging
Pending clean PR	Mailing list consensus call on Top N Tracks in the base draft
14 November, San Francisco	IETF 127

One correction worth making loudly, because we got it wrong ourselves in the IETF 126 recap and have since fixed it: when the working group said implementations should be on draft-20 “for the Seattle meeting”, that means the in-person interim on 14–15 October, not the next IETF. IETF 127 is in San Francisco in November. The draft-20 deadline is a month earlier than the plenary calendar suggests.

Update, 22 August 2026. The 10 August interim revised this release plan. Draft-20 is now an editorial cut taken just before the in-person editors’ meeting, draft-21 captures that meeting’s output, and draft-22 is the working group’s next official interop target — with no date attached. The 14–15 October Seattle date stands; the revision you are expected to arrive running does not. See Three Meetings to Change a Filter.

How to read forty-four drafts

If you are trying to work out what to actually track, the honest answer depends on what you are building:

Implementing MoQT? transport-19, plus the filter work (PR 1809, PR 1830) and SSTS. Everything else can wait for last call.
Shipping media over it? MSF and CMSF, then LOC or LOCMAF depending on whether you have a CMAF pipeline. Add the SCTE-35 draft if you monetize.
Running relays? The relay operations cluster, plus qlog, plus the diagnostics document when it appears.
Deploying in a regulated or multi-tenant environment? C4M and Secure Objects, and watch the Seattle interim on 14–15 October.
Building agents? Watch agentproto’s chartering. Do not build on any of those drafts yet.
Just want the shape of it? draft-westerlund-moq-overview-00, then our MoQT explainers.

Next issue in two weeks, after the 24 August interim. We should have the September interop day date, movement on the draft-20 PR queue, and quite possibly a Top N Tracks consensus call to report on.

MoQ News #1: Seven Adopted Drafts, Forty-Four Active Ones, and a Map of the Pile