Alternatives
Products that do what Keep large tool output out of LLM context: 3x accuracy 95% fewer tokens does
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…
- 1WW
I spent a few hours last weekend testing whether AI can replace code by executing directly. Built a contact manager where every HTTP request goes to an LLM with three tools: database (SQLite), webResponse (HTML/JSON/JS), and updateMemory (feedback). No routes, no controllers, no business logic. The AI designs schemas on first request, generates UIs from paths alone, and evolves based on natural language feedback. It works—forms submit, data persists, APIs return JSON—but it's catastrophically slow (30-60s per request), absurdly expensive ($0.05/request), and has zero UI…
Nov 2025 · github.com
- 2AJ
Hey HN, we’re building an open specification that lets agents discover and invoke APIs with natural language, built on the OpenAPI standard. agents.json clearly defines the contract between LLMs and API as a standard that's open, observable, and replicable. Here’s a walkthrough of how it works: https://youtu.be/kby2Wdt2Dtk?si=59xGCDy48Zzwr7ND. There’s 2 parts to this: 1. An agents.json file describes how to link API calls together into outcome-based tools for LLMs. This file sits alongside an OpenAPI file. 2. The agents.json SDK loads agents.json files as tools for an LLM that…
2025 · github.com
- 3AK
I shipped a wiki layer for AI agents that uses markdown + git as the source of truth, with a bleve (BM25) + SQLite index on top. No vector or graph db yet. It runs locally in ~/.wuphf/wiki/ and you can git clone it out if you want to take your knowledge with you. The shape is the one Karpathy has been circling for a while: an LLM-native knowledge substrate that agents both read from and write into, so context compounds across sessions rather than getting re-pasted every morning. Most implementations of that idea land on Postgres, pgvector, Neo4j, Kafka, and a dashboard. I…
Apr 2026 · github.com
- 4
- 5KR
I discovered that in LLM inference, keys and values in the KV cache have very different quantization sensitivities. Keys need higher precision than values to maintain quality. I patched llama.cpp to enable different bit-widths for keys vs. values on Apple Silicon. The results are surprising: - K8V4 (8-bit keys, 4-bit values): 59% memory reduction with only 0.86% perplexity loss - K4V8 (4-bit keys, 8-bit values): 59% memory reduction but 6.06% perplexity loss - The configurations use the same number of bits, but K8V4 is 7× better for quality This means you can run LLMs with 2-3× longer…
2025 · github.com
- 6KO
We've open-sourced Klarity - a tool for analyzing uncertainty and decision-making in LLM token generation. It provides structured insights into how models choose tokens and where they show uncertainty. What Klarity does: - Real-time analysis of model uncertainty during generation - Dual analysis combining log probabilities and semantic understanding - Structured JSON output with actionable insights - Fully self-hostable with customizable analysis models The tool works by analyzing each step of text generation and returns a structured JSON: - uncertainty_points: array of {step, entropy,…
2025 · github.com
- 7
- 8RL
We've been building data pipelines that scrape websites and extract structured data for a while now. If you've done this, you know the drill: you write CSS selectors, the site changes its layout, everything breaks at 2am, and you spend your morning rewriting parsers. LLMs seemed like the obvious fix — just throw the HTML at GPT and ask for JSON. Except in practice, it's more painful than that: - Raw HTML is full of nav bars, footers, and tracking junk that eats your token budget. A typical product page is 80% noise. - LLMs return malformed JSON more often than you'd expect, especially with…
Mar 2026 · github.com
- 9MO
Every MCP server injects its full tool schemas into context on every turn — 30 tools costs ~3,600 tokens/turn whether the model uses them or not. Over 25 turns with 120 tools, that's 362,000 tokens just for schemas. mcp2cli turns any MCP server or OpenAPI spec into a CLI at runtime. The LLM discovers tools on demand: mcp2cli --mcp https://mcp.example.com/sse --list # ~16 tokens/tool mcp2cli --mcp https://mcp.example.com/sse create-task --help # ~120 tokens, once mcp2cli --mcp https://mcp.example.com/sse create-task --title "Fix bug" No…
Mar 2026 · github.com
- 10AT
I recently built a small open-source tool to benchmark different LLM API endpoints — including OpenAI, Claude, and self-hosted models (like llama.cpp). It runs a configurable number of test requests and reports two key metrics: • First-token latency (ms): How long it takes for the first token to appear • Output speed (tokens/sec): Overall output fluency Demo: https://llmapitest.com/ Code: https://github.com/qjr87/llm-api-test The goal is to provide a simple, visual, and reproducible way to evaluate performance across different LLM providers, including…
2025 · llmapitest.com
- 11AN
When building workflows that rely on LLMs, we commonly use structured output for programmatic use cases like converting an invoice into rows or meeting transcripts into tickets or even complex PDFs into database entries. The model may return the schema you want, but with hallucinated values like `invoice_date` being off by 2 months or the transcript array ordered wrongly. The JSON is valid, but the values are not. Structured output today is a big part of using LLMs, especially when building deterministic workflows. Current structured output benchmarks (e.g., JSONSchemaBench) only validate…
Apr 2026 · interfaze.ai
- 12
- 13CT
If agent's tools are exposed as functions/objects in a Python REPL (as opposed to JSON schemas) they perform better, I linked the explainer article we wrote, but if you want to jump straight in check out the docs! https://docs.symbolica.ai/
Dec 2025 · symbolica.ai
- 14FA
LLM agents rely on tool calls — but tool responses are huge. Gmail, CRMs, and APIs return bloated JSON LLMs choke on large responses You only need 2–3 fields, but frameworks give you zero control Toolflow is an AI-native framework to fix this: * Filter tool responses before they hit the LLM * Context modes: `minimal`, `full`, `custom`, or `ai` * Composable TypeScript tool registry GitHub: [https://github.com/dksingh1997/toolflow](https://github.com/dksingh1997/toolflow) Would love feedback — especially from those building with LLMs in production.
2025 · github.com
- 15
- 16TC
Hi HN, I built a CLI for uploading documents and querying them with an LLM agent that uses search tools rather than stuffing everything into the context window. I recorded a demo using the CrossFit 2025 rulebook that shows how this approach compares to traditional RAG and direct context injection[1]. The core insight is that LLMs running in loops with tool access are unreasonably effective at this kind of knowledge retrieval task[2]. Instead of hoping the right chunks make it into your context, the agent can iteratively search, refine queries, and reason about what it finds. The CLI handles…
2025 · github.com
- 17PR
Hi HN, While building RAG agents, I noticed a lot of token budget was wasted on formatting overhead (HTML tags, JSON structure, whitespace). Existing solutions felt too heavy (often requiring torch/transformers), so I wrote this lightweight, zero-dependency library to solve it. It includes strategies for context packing, PII redaction, and tool output compression. Benchmarks show it can save ~15% of tokens with negligible latency overhead (<0.5ms). Happy to answer any questions!
Dec 2025 · github.com
- 18OS
Hi HN, Matvey, Ildar, Joey, and Dominik here. If you're building LLM agents that use tools, you're probably worried about prompt injection attacks that can hijack those tools. We were too, and found that solutions like prompt-based filtering or secondary "guard" LLMs can be unreliable. Our thesis is that agent security should be handled at the network level between the agent and the LLM, just like a traditional web application firewall. So we built Archestra Platform: an open-source gateway that acts as a secure proxy for your AI agents. It's designed to be a deterministic firewall against…
Oct 2025 · archestra.ai
- 19MG
Many teams connecting LLMs to external tools eventually encounter the same architectural issue: as more tools and agents are added, the integration pattern becomes an N×M mesh of direct connections. Each agent implements its own auth, retries, rate limiting, and logging; each tool needs credentials distributed to multiple places and observability becomes fragmented. We built LLM gateway with this goal to provide a single place to manage authentication, authorization, routing, and observability for MCP servers, with a path toward a more general agent-gateway architecture in the future. The…
Dec 2025 · truefoundry.com
- 20RL
While working with LLMs for structured web data extraction, we saw issues with invalid JSON and broken links in the output. This led me to build a library focused on robust extraction and enrichment: - Clean HTML conversion: transforms HTML into LLM-friendly markdown with an option to extract just the main content - LLM structured output: Uses Gemini 2.5 flash or GPT-4o mini to balance accuracy and cost. Can also also use custom prompt - JSON sanitization: If the LLM structured output fails or doesn't fully match your schema, a sanitization process attempts to recover and fix the data,…
2025 · github.com
- 21
Token-efficiency linter for LLM prompts and payloads - ritenv/tokensift
8d ago · github.com
- 22MS
Hey HN, I’m the author. I built Misata because existing tools (Faker, Mimesis) are great for random rows but terrible for relational or temporal integrity. I needed to generate data for a dashboard where "Timesheets" must happen after "Project Start Date," and I wanted to define these rules via natural language. How it works: LLM Layer: Uses Groq/Llama-3.3 to parse a "story" into a JSON schema constraint config. Simulation Layer: Uses Vectorized NumPy (no loops) to generate data. It builds a DAG of tables to ensure parent rows exist before child rows (referential integrity).…
Dec 2025 · github.com
- 23AC
Multi-tier exact-match cache for AI agents backed by Valkey or Redis. LLM responses, tool results, and session state behind one connection. Framework adapters for LangChain, LangGraph, and Vercel AI SDK. OpenTelemetry and Prometheus built in. No modules required - works on vanilla Valkey 7+ and Redis 6.2+. Shipped v0.1.0 yesterday, v0.2.0 today with cluster mode. Streaming support coming next. Existing options locked you into one tier (LangChain = LLM only, LangGraph = state only) or one framework. This solves both. npm:…
Apr 2026
- 24
Ranked by how close each launch is in meaning, then by votes. Refine with a description →