frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Keep large tool output out of LLM context: 3x accuracy 95% fewer tokens

https://github.com/lourencomaciel/sift-gateway
6•loumaciel•4h ago
LLM agents often place raw JSON tool outputs directly in the prompt. After a few tool calls, earlier results get compacted or truncated and answers become incorrect or inconsistent.

I built Sift, a drop-in MCP gateway that stores tool outputs as local artifacts (filesystem blobs indexed in SQLite) and returns an `artifact_id` plus compact schema hints when responses are large or paginated.

Instead of reasoning over full JSON in the prompt, the model runs a small Python query:

    def run(data, schema, params):
        return max(data, key=lambda x: x["magnitude"])["place"]
Query code runs in a constrained subprocess (AST/import guards + timeout/memory caps). Only the computed result is returned to the model.

Benchmark (Claude Sonnet 4.6, 103 questions across 12 datasets):

- Baseline (raw JSON in prompt): 34/103 (33%), 10.7M input tokens

- Sift (artifact + code query): 102/103 (99%), 489K input tokens

Open benchmark + MIT code: https://github.com/lourencomaciel/sift-gateway

Install:

    pipx install sift-gateway
    sift-gateway init --from claude
Works with Claude Code, Cursor, Windsurf, Zed, and VS Code. Existing MCP servers and tools require no changes.

Comments

loumaciel•4h ago
Happy to answer questions about the sandboxing, artifact format, or the benchmark setup.

The benchmark harness and datasets are in the repo if anyone wants to reproduce or extend the tests. Curious if others have run into the same context compaction issues with tool-heavy agents.

Show HN: Jido 2.0, Elixir Agent Framework

https://jido.run/blog/jido-2-0-is-here
119•mikehostetler•2h ago•31 comments

Show HN: PageAgent, A GUI agent that lives inside your web app

https://alibaba.github.io/page-agent/
16•simon_luv_pho•1h ago•4 comments

Show HN: Poppy – A simple app to stay intentional with relationships

https://poppy-connection-keeper.netlify.app/
162•mahirhiro•14h ago•75 comments

Show HN: Tracemap – run and visualize traceroutes from probes around the world

https://tracemap.dev/
4•solhuang•1h ago•2 comments

Show HN: OmoiOS–190K lines of Python to stop babysitting AI agents (Apache 2.0)

https://github.com/kivo360/OmoiOS
2•kanddle•2h ago•2 comments

Show HN: AgnosticUI – A source-first UI library built with Lit

https://www.agnosticui.com/
3•roblevintennis•2h ago•1 comments

Show HN: echo.html, between Feather Wiki and Roam with commands like Emacs

https://m15o.net/echo/
3•m15o•3h ago•0 comments

Show HN: Hormuz Crisis Dashboard Real-time shipping disruption tracker

https://www.hormuztracker.com/
4•MrNekked•3h ago•0 comments

Show HN: Stacked Game of Life

https://stacked-game-of-life.koenvangilst.nl/
188•vnglst•5d ago•26 comments

Show HN: Keep large tool output out of LLM context: 3x accuracy 95% fewer tokens

https://github.com/lourencomaciel/sift-gateway
6•loumaciel•4h ago•1 comments

Show HN: Voice skill for AI agents – sub-200ms latency via native SIP

https://github.com/nia-agent-cyber/openai-voice-skill
2•nia-agent•5h ago•0 comments

Show HN: SpiderSuite – Multi-engine web crawler and proxy for security research

https://spidersuite.io/
3•sub3suite•5h ago•1 comments

Show HN: Vertex.js – A 1kloc SPA Framework

https://lukeb42.github.io/vertex-manual.html
43•LukeB42•4d ago•25 comments

Show HN: A shell-native cd-compatible directory jumper using power-law frecency

https://github.com/jghub/sd-switchdir
23•jghub•1d ago•7 comments

Show HN: Rust compiler in PHP emitting x86-64 executables

https://github.com/mrconter1/rustc-php
64•mrconter11•4d ago•48 comments

Show HN: podcast-cli - A Rust CLI for Podcast Index & YouTube Subtitles

https://github.com/the-waste-land/podcast-cli
2•liweixin•6h ago•1 comments

Show HN: DevTrack – A personal dashboard to track your developer growth

https://devtrack-rose.vercel.app
3•nullAffi•6h ago•0 comments

Show HN: Anaya – CLI that scans codebases for DPDP compliance violations

https://github.com/sandip-pathe/anaya-scan
4•sandippathe•7h ago•1 comments

Show HN: I made a zero-copy coroutine tracer to find my scheduler's lost wakeups

https://github.com/lixiasky-back/coroTracer
44•lixiasky•2d ago•3 comments

Show HN: AlifZetta – AI Operating System That Runs LLMs Without GPUs

https://axz.si/
4•padamkafle•8h ago•1 comments

Show HN: PyMath Preview – preview LaTeX math in Python docstrings inside VS Code

https://github.com/sankarebarri/pymath-preview
2•sankarebarri•8h ago•1 comments

Show HN: Your AI Slop Bores Me

https://www.youraislopbores.me/
11•mikidoodle•11h ago•4 comments

Show HN: A GFM+GF-MathJax/Latex HTML formatting adventure

https://github.com/scottvr/phart/blob/main/docs/GHM-LATEX.md
4•ycombiredd•4d ago•1 comments

Show HN: Paste a URL and watch multiple AI models redesign it side-by-side

https://shuffle.dev/ai-website-redesign
7•kemyd•19h ago•0 comments

Show HN: I built a sub-500ms latency voice agent from scratch

https://www.ntik.me/posts/voice-agent
564•nicktikhonov•2d ago•153 comments

Show HN: Timber – Ollama for classical ML models, 336x faster than Python

https://github.com/kossisoroyce/timber
204•kossisoroyce•3d ago•33 comments

Show HN: Omni – Open-source workplace search and chat, built on Postgres

https://github.com/getomnico/omni
172•prvnsmpth•3d ago•42 comments

Show HN: Qlog – grep for logs, but 100x faster

https://github.com/Cosm00/qlog
15•cosm00•21h ago•17 comments

Show HN: I put HN discussions next to the article where it belongs

https://cool-link-web-production.up.railway.app/l/blogabout
8•krenerd•1d ago•0 comments

Show HN: Fast Chladni figure simulation in Python with NumPy vectorization

https://github.com/ratwolfzero/Chladni_Figures
4•ratwolf•13h ago•1 comments