ARTF Containers

Experts in ARTF and adtech containerization

The IAB Tech Lab's Agentic Real Time Framework (ARTF) standardizes something we have already shipped: adtech workloads packaged as containers and deployed inside a host platform's infrastructure, next to the auction. The spec defines container runtime requirements and an API for reliable, protected and private bidstream mutation. Package once, deploy to any compliant platform.

We work on both sides of that sentence. We containerize adtech stacks, from full bidders down to single-purpose service agents, and we track the ARTF specification as it moves from version 1.0 (public comment closed January 2026) toward version 2.0 in the Agentic Working Group, whose progress we follow alongside our buyer agent and seller agent work.

The first containerized DSP, built end to end

Bedrock Platform is the next-generation DSP, containerized from day one. Appliscale is Bedrock's investor and engineering partner: we built the platform end to end and continue to ship week over week.

This is not a prototype. Bedrock's bidders are live on exchange traffic today, handling real auctions at production volume. The full build is documented in our Bedrock Platform case study.

What the Agentic Real Time Framework standardizes

Before ARTF, every host platform that wanted third-party processing in its bidstream negotiated a custom integration, and every provider rebuilt its packaging for each host. ARTF replaces that with one contract:

The use cases go beyond bidding: identity resolution, fraud detection and deal activation all fit the same model of a service agent operating within a host platform. A bidder is simply the hardest of them, which is why having built one is the strongest possible reference.

ARTF on the wire: gRPC and protobuf, measured

ARTF replaces programmatic's JSON-over-HTTP with gRPC and protobuf, and our Head of Engineering Damian Naglak, who works hands-on with Bedrock, has published the reasoning and the measurements. His breakdown of why ARTF runs on gRPC and protobuf walks the trade-offs for an exchange handling tens of billions of bid requests a day. Then he benchmarked it end to end with 300,000 real bid requests sampled from our own bidstream, TLS and gzip on both sides: protobuf came out about 19 percent smaller on the wire after gzip and used about 15 percent less CPU, and gRPC's connection multiplexing kept median latency 30 to 40 percent lower under load while absorbing about a quarter more throughput on the same hardware. On a bid path, a response that misses the auction window never bid.

The market is moving the same direction. At Cannes Lions 2026, containerized bidding was the topic people raised unprompted, from supply platforms and premium video to audio and infrastructure providers.

Why containerized bidstream processing changes the economics

Classic RTB sends every bid request across the public internet from SSP to DSP and back. A containerized deployment inverts that: the logic runs inside the host's environment, next to the auction.

Two consequences follow directly. First, auction data never leaves the host, because the container executes where the data already lives. Second, egress costs collapse, because cloud providers bill on outgoing traffic and the hot path no longer crosses a metered boundary.

The cloud vendors see the same shift. AWS RTB Fabric claims single-digit millisecond latency over a dedicated network and up to 80 percent savings versus standard data transfer fees. We analysed those claims in public, caveats included, in our AWS RTB Fabric deep dive.

ARTF and containerization services

This work sits inside our broader advertising technology practice, alongside seller agent development and Prebid trusted server work. For investors evaluating adtech infrastructure plays, see how we work with tech investors.

Engineering proof at auction scale

Containerized or not, a bidder is judged on the hot path. The numbers behind Bedrock:

The containerized deployment handles half a million queries per second and runs inside IndexCloud, Index Exchange's proprietary datacenter, colocated with the exchange itself. Around it sit a VictoriaMetrics and Grafana observability stack and GitHub Actions CI/CD, keeping responses under 100 milliseconds.

The scale ceiling is higher than Bedrock. Our engineers optimized Yahoo's RTB platform at up to 1.5 million bid requests per second across four global regions. We work with AdCP, IAB standards and OpenRTB in production.

Deep dives from the team

We write about this work as we do it. The Containerized Bidder series is an architectural deep dive into a bidder shipping in containers: the runtime contract, data access under latency budgets, and what the host environment actually expects.

The AWS RTB Fabric analysis walks the cost model line by line: when the 80 percent savings claim holds, when colocation already beats it, and who should adopt it first.