retro · architecture · liquidsoap · history
How the Heck Did We Get Here? 5 Years of Visual Scripting for Liquidsoap
In software development, it is easy to look at a finished product backward and assume the architectural plan was always obvious. For MediaMesh, the core vision was established five years ago and never changed: build a visual node-based editor over Liquidsoap that renders directly to real, runnable source code.
What actually took five years wasn’t refining the vision — it was waiting for the surrounding tooling, AI ecosystems, and protocol infrastructure to catch up to what we wanted to build.
This is the story of how MediaMesh got here.
1. The Liquidshop Era (2021)
Long before MediaMesh v2, the foundational ideas were publicly presented during Liquidshop 2 — Liquidsoap’s official community conference.
Chris Mahoney and Rob Carballo gave a talk titled “MediaMesh: An Approach to Complex Media Productions”, laying out the exact thesis that drives the project today: visual graphs representing real broadcast pipelines rather than simplified abstractions.
A Handle on Infrastructure, A Wall on the Catalog
The Liquidshop Era had real command over infrastructure and orchestration — the hard, unglamorous plumbing that the current system still stands on.
What stopped the early iteration cold was trying to hand-transcribe Liquidsoap’s operator catalog. Attempting to manually maintain hundreds of Liquidsoap function signatures one by one created an unmaintainable maintenance wall whenever the upstream binary changed.
MediaMesh v2 solved this fundamental problem by abandoning hand-written catalog definitions entirely. Instead, the modern engine dynamically extracts and normalizes the operator catalog directly from the compiler itself.
2. The Original R&D Goal: June 2021
TrueMotive’s initial R&D specification in June 2021 targeted one explicit output:
- A visual node-canvas interface over Liquidsoap.
- Every node and wire maps directly to valid Liquidsoap code.
- Checked and compiled by the native Liquidsoap binary before execution.
The catalog extraction model, the compile-before-save pipeline, and the canvas builder were all part of the original 2021 specification. The goal was correct on day one — the challenge was building the precondition tooling to support it at speed.
3. The Funding & Runway Reality
Being right early doesn’t pay the bills of being right early.
The team maintained unusual clarity on where cloud streaming architecture was headed, but converting a five-year technical vision into immediate runway as a small team is a hard reality in software development.
A small team with a correct five-year plan looks remarkably similar on paper to a small team with an incorrect one until the surrounding ecosystem matures.
4. Waiting for Ecosystem Convergence
During this period, TrueMotive was accepted into the Microsoft for Startups program, gaining early access to OpenAI APIs and cloud development credits.
Rather than spending those resources on monolithic prototypes, every unit of effort went toward dependency isolation:
- Structurally separating the streaming engine, web console, protocol proxies (
MediaMTX), and control servers into independent micro-containers sharing a unified filesystem. - Keeping the visual compiler decoupled from runtime state.
The Three Missing Ecosystems
For five years, three distinct technical ecosystems had to evolve and converge before MediaMesh could reach its full potential:
- Mature Media Tooling: Off-the-shelf sidecars (MediaMTX for SRT/WHEP/RTMP) and Liquidsoap’s introspectable AST catalog.
- Capable AI Assistants: AI tools mature enough to operate within a complex modular architecture via protocol standards like MCP (Model Context Protocol).
- Container & Network Infrastructure: Zero-Trust VPN mesh networking (Tailscale), Durable Objects for state, and standard Docker orchestration.
Once all three ecosystems converged simultaneously, what once required months of manual plumbing could be built, validated, and shipped in days.
5. Retrospective Verdict
Holding a specific, unfashionable-at-the-time vision for five straight years, hitting a technical wall early, and being vindicated when the surrounding ecosystem catches up is a rare trajectory in product engineering.
The vision didn’t change — the world around it finally arrived.