tokens on Hacker News

66 titles from 63 authors in the last 30 days, with 61 distinct links and 19 Show HN.

Counts titles, it does not confirm facts. Open the source before concluding anything.

Projects on this term

The score comes from our model.

• Show HN: How many tokens are you wasting in your context window? · score 7.8

  context-autopsy analyzes local coding-agent transcripts to break down where tokens are spent in the context window, across many agents like Claude Code and Cursor. It distinguishes one-time token additions from 'processed' tokens (multiplied by model calls), highlighting waste from build logs, failed commands, and repeated content. The project provides actionable insights and shareable summaries to help developers optimize their agent usage and costs.

• Show HN: Plain-text code memory that beats a graph DB at x218 fewer tokens · score 7.5

  DaiDocs introduces a plain-text memory format (.dai for documents/history, .cai for code graphs) that enables AI assistants to store and retrieve context efficiently, beating graph databases with dramatically fewer tokens. It combines a deterministic code graph via tree-sitter with an LLM-based observer for documents, all stored locally as grep-able files. The project provides rigorous benchmarks and a router to direct queries to the appropriate memory tier.

• Show HN: Mixdog – stretch your Claude/ChatGPT subscription with 63% fewer tokens · score 7.3

  Mixdog is an open-source AI coding agent that reduces token usage by up to 63% compared to Codex CLI while maintaining the same success rate, using techniques like auto reasoning effort and context efficiency. It supports multiple LLM providers (Claude, ChatGPT, Grok, local models), offers a desktop app and CLI, and includes advanced features like swarm mode, bots, and browser/desktop control.

• Show HN: I Cut my Agent Tokens By 50% · score 7.3

  indexio is a Rust-based code search engine designed specifically for AI coding agents, indexing entire organizations' repositories once and serving fast lexical, semantic, and hybrid queries via MCP and HTTP. It combines trigram indexing, vector search, and code graph analysis to reduce token usage for agents like Claude Code and Cursor.

Recent titles

1. Show HN: Universal Tokens – like Universal Paperclips, but tokens
   2026-11-02 · mani.wtf · HN discussion
2. Show HN: Kvark Sparkle 27B – reasoning model trained to think with fewer tokens
   2026-11-02 · huggingface.co · HN discussion
3. Show HN: Knock – stop Claude Code burning tokens exploring big codebases
   2026-11-02 · github.com · HN discussion
4. The tokens in your agent harness
   2026-11-01 · armincat.com · HN discussion
5. Codedict – a no-AI code index that cut my Claude Code tokens by 63%
   2026-11-01 · github.com · HN discussion
6. Cisco gives engineer unlimited AI tokens to rewrite company's legacy code
   2026-10-31 · tomshardware.com · HN discussion
7. Non-autoregressive NC LLM spits tokens 21x faster than text diffusion
   2026-10-31 · hazyresearch.stanford.edu · HN discussion
8. Same agent-memory accuracy with 8x fewer tokens by changing the embedding model
   2026-10-30 · github.com · HN discussion
9. Binance hot wallet drained of about $2.5B in tokens on Ethereum
   2026-10-29 · etherscan.io · HN discussion
10. If the context window is a million tokens, why bother with search?
    2026-10-29 · deployed-ai.io · HN discussion
11. Show HN: Endlesstokens – an LLM API billed by time, not tokens
    2026-10-29 · endlesstokens.ai · HN discussion
12. Claude Code burns ~15–20% of your tokens on a 3-word status line
    2026-10-29 · gist.github.com · HN discussion
13. How many tokens does a Claude subscription buy?
    2026-10-28 · claudesubscriptiontokencount.com · HN discussion
14. Show HN: Stranded, move ERC-20 tokens out of a wallet that has no ETH for gas
    2026-10-27 · strandedtoken.trade · HN discussion
15. AuthMask: Microsoft Authenticator hands tokens to an app with zero permissions
    2026-10-27 · enclave.ai · HN discussion
16. RFC AI Tooling: 40x fewer tokens while reading the RFC corpus
    2026-10-27 · github.com · HN discussion
17. OpenClaw creator says he spent $10M in tokens last month
    2026-10-26 · theregister.com · HN discussion
18. Best engineers should get unlimited tokens
    2026-10-24 · keithbrown.com · HN discussion
19. Ask HN: What was that repo I can donate/sell me remaining subscription tokens?
    2026-10-24 · HN discussion
20. Show HN: Mdt – save tokens reading changes to Markdown
    2026-10-23 · github.com · HN discussion
21. Evervault Browser Tokens: Agentic payments on any website
    2026-10-23 · evervault.com · HN discussion
22. Another approach to save tokens from coding agents
    2026-10-22 · github.com · HN discussion
23. Show HN: Blastradius – find PHP2Shell, Node, emSign or leaked tokens
    2026-10-22 · github.com · HN discussion
24. Show HN: Pith – a proxy that cuts LLM output tokens, and refuses when it can't
    2026-10-22 · github.com · HN discussion
25. Canary Tokens
    2026-10-21 · canarytokens.org · HN discussion
26. How to save money on tokens for your AI agents
    2026-10-20 · terraphim.ai · HN discussion
27. Asking Astra to decipher the Indus script: 891M tokens later
    2026-10-19 · chinmay.ai · HN discussion
28. Hybrid Reasoning Agents: 35–55% fewer tokens and up to 2.6× faster inference
    2026-10-18 · arxiv.org · HN discussion
29. For those using huge amount of tokens monthly for dev work, why do you need it?
    2026-10-17 · HN discussion
30. Scaling Indic Pretraining: 2.8T Tokens and an Open Data Generation Pipeline
    2026-10-17 · huggingface.co · HN discussion

All trends