frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Tessera – An open protocol for AI-to-AI knowledge transfer

https://github.com/incocreativedev/tessera-core
3•kirkmaddocks•2h ago
Tessera is an activation-based protocol that lets trained ML models transfer knowledge to other models across architectures. Instead of dumping weight tensors, it encodes what a model has learnt — activations, feature representations, behavioural patterns — into self-describing tokens that a receiving model can decode into its own architecture.

The reference implementation (tessera-core) is a Python/PyTorch library. Current benchmarks show positive transfer across CNN, Transformer, and LSTM pairs. It runs on CPU and the demo finishes in under 60 seconds.

Happy to answer questions about the protocol design, the wire format, or the benchmark methodology.

Comments

0xecro1•2h ago
Interesting approach. I work in embedded Linux/edge AI where we constantly struggle to move knowledge from large training models down to quantized INT8 models on constrained hardware (ARM Cortex-A class). Have you tested transfer to quantized or pruned targets? If the behavioural encoding survives that compression, this could be a much cleaner path than classical distillation for on-device deployment.
kirkmaddocks•2h ago
We haven't built quantisation-aware transfer yet, but the architecture lends itself to it better than you might expect.

Mode A (activation transfer) operates at the representation level, not the parameter level. The source model's knowledge gets projected into a 2048-dim hub space — the receiving model doesn't need to match architecturally or in precision. A 200M FP32 training model and a 5M INT8 edge model can both have UHS encoders/decoders. The hub space is agnostic to what's underneath.

Mode B (behavioural) is probably the most interesting path for your use case. It transfers decision boundaries rather than activations or weights. If the quantised model can reproduce the input-output mapping, internal precision is irrelevant.

It's similar in spirit to distillation but decoupled through the hub space — teacher and student don't need to be online simultaneously, and you get a full audit trail of what knowledge went where (which matters if you're shipping medical/industrial edge models under EU AI Act).

The gap today is the decoder side. DecoderMLP outputs FP32. We'd need a quantisation-aware variant that respects the INT8 grid — straight-through estimator at minimum, learned quantisation boundaries ideally. We'd also want empirical drift characterisation across FP32→FP16→INT8→INT4 so you'd know your expected fidelity floor for a given target.

The swarm angle is where it gets genuinely useful for edge fleets. If you've got N devices training locally on-site data, they contribute quantised-model tokens back to a full-precision aggregator. The robust aggregation strategy (Huber-style cosine clipping) handles quantisation noise across heterogeneous devices naturally.

We're planning a quantisation-aware transfer module next. If you're interested in testing against real Cortex-A INT8 workloads, we'd welcome the collaboration — repo is at github.com/incocreativedev/tessera-core.

Show HN: enveil – hide your .env secrets from prAIng eyes

https://github.com/GreatScott/enveil
154•parkaboy•10h ago•89 comments

Show HN: X86CSS – An x86 CPU emulator written in CSS

https://lyra.horse/x86css/
199•rebane2001•13h ago•67 comments

Show HN: Steerling-8B, a language model that can explain any token it generates

https://www.guidelabs.ai/post/steerling-8b-base-model-release/
255•adebayoj•15h ago•75 comments

Show HN: Awsim – Lightweight AWS emulator in Go (40 services in progress)

https://github.com/sivchari/awsim
2•sivchari•1h ago•0 comments

Show HN: PgDog – Scale Postgres without changing the app

https://github.com/pgdogdev/pgdog
295•levkk•1d ago•54 comments

Show HN: AI-native SDLC – 156 test docs, 16 skills, 1 human

https://github.com/gpgkd906/auth9
3•gpgkd906•1h ago•0 comments

Show HN: Dicta.to – Local voice dictation for Mac with on-device AI

https://dicta.to/
2•alamparelli•1h ago•0 comments

Show HN: AI phone assistant that became a lifeline for people who can't speak

https://mio.gg/
3•talyuk•1h ago•1 comments

Show HN: Out Plane – A PaaS I built solo from Istanbul in 3 months

https://outplane.com
4•receperdogan•4h ago•2 comments

Show HN: Cellarium: A Playground for Cellular Automata

https://github.com/andrewosh/cellarium
33•andrewosh•3d ago•0 comments

Show HN: Babyshark – Wireshark made easy (terminal UI for PCAPs)

https://github.com/vignesh07/babyshark
131•eigen-vector•18h ago•43 comments

Show HN: Tessera – An open protocol for AI-to-AI knowledge transfer

https://github.com/incocreativedev/tessera-core
3•kirkmaddocks•2h ago•2 comments

Show HN: Sowbot – Open-hardware agricultural robot (ROS2, RTK GPS)

https://sowbot.co.uk/
168•Sabrees•23h ago•46 comments

Show HN: WebPerceptor – Enabling AI Mediated Web Browsing

https://github.com/theartofhci/WebPerceptor
2•theartofhci•2h ago•0 comments

Show HN: Claude Copy – Drop-in fix for Claude Code's broken copy-paste

https://github.com/andersmyrmel/claude-copy
2•andersmyrmel•2h ago•0 comments

Show HN: AI Timeline – 171 LLMs from Transformer (2017) to GPT-5.3 (2026)

https://llm-timeline.com/
163•ai_bot•1d ago•56 comments

Show HN: Git-native-issue – issues stored as commits in refs/issues/

https://github.com/remenoscodes/git-native-issue
4•remenoscodes•4h ago•3 comments

Show HN: CIA World Factbook Archive (1990–2025), searchable and exportable

https://cia-factbook-archive.fly.dev/
482•MilkMp•1d ago•99 comments

Show HN: L88 – A Local RAG System on 8GB VRAM (Need Architecture Feedback)

7•adithyadrdo•10h ago•1 comments

Show HN: Notion-CLI – Full Notion API from the terminal, 39 commands, one binary

https://github.com/4ier/notion-cli
3•4ier•8h ago•1 comments

Show HN: 3D Mahjong, Built in CSS

https://voxjong.com
133•rofko•1d ago•60 comments

Show HN: 60 Years of Metal Music Data, Visualized

https://metal-archives-graphs.neocities.org
3•oneyoudontknow•3h ago•0 comments

Show HN: Agent Multiplexer – manage Claude Code via tmux

https://github.com/mixpeek/amux
12•Beefin•23h ago•2 comments

Show HN: AgentBudget – Real-time dollar budgets for AI agents

https://github.com/sahiljagtap08/agentbudget
6•sahiljagtapyc•9h ago•6 comments

Show HN: A geometric analysis of Chopin's Prelude No. 4 using 3D topology

https://github.com/jimishol/cholidean-harmony-structure/blob/main/docs/03-case-study-chopin-prelu...
48•jimishol•4d ago•11 comments

Show HN: BVisor – An Embedded Bash Sandbox, 2ms Boot, Written in Zig

https://github.com/butter-dot-dev/bVisor
18•edunteman•21h ago•5 comments

Show HN: ClinTrialFinder –AI-powered clinical trial matching for cancer patients

https://www.clintrialfinder.info
2•chncwang•10h ago•0 comments

Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU

https://github.com/xaskasdf/ntransformer
388•xaskasdf•2d ago•102 comments

Show HN: Local-First Linux MicroVMs for macOS

https://shuru.run
207•harshdoesdev•1d ago•63 comments

Show HN: Implementing ping from the Ethernet layer (ARP,IPv4,ICMP in user space)

https://github.com/v420v/ping
8•ibuki256•1d ago•1 comments