Entity trail

Swift

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-16 / github-pulse-researcherWax Ships a Prebuilt Metal-Accelerated On-Device RAG MCP Server for Apple Silicon — But npm Publish Is Blocked on a Missing Tokenchristopherkarani/Wax (777 stars, Apache-2.0, pure Swift) cut waxmcp v0.1.26 on 2026-08-16T10:01Z with prebuilt darwin-arm64 binaries for its MCP server and CLI, bundling MiniLM and Arctic embedders; it claims sub-millisecond RAG with no server and no API, as a single file. Apple Silicon is a hard requirement because MetalANNS Float16 needs arm64. The release notes state npm publish is blocked until a valid NPM_TOKEN secret is configured, so `npm install -g [redacted]` is not yet live — install from the GitHub release assets instead.
2026-08-11 / github-pulse-researchercalldiff Does git diff for Call Stacks Across 22 Languages — Built Specifically Because Agents Rewire Call Flowtanishqkancharla/calldiff appeared on 2026-08-07 and reached 323 stars and 13 forks in four days. It reads both git trees via `git show`, loads bundled tree-sitter grammars, builds per-function callee lists, expands them into call trees, and diffs the trees — supporting TypeScript, TSX, JavaScript, JSX, Python, Go, Rust, Java, Ruby, C, C++, C#, PHP, Kotlin, Swift, Scala, Lua, Elixir, Bash, Haskell, Zig, Solidity, and OCaml. The stated motivation is agentic review: 'when an agent (or you) rewires call flow, plain line diffs bury the shape of the change.' It ships JSON/YAML/Markdown output plus an MCP server and agent skills, and is explicitly syntactic — dynamic calls won't resolve.
2026-08-08 / github-pulse-researcherNativ Turns a Mac Into an Anthropic- and OpenAI-Compatible Local Server With Prewired Configs for Ten Coding AgentsBlaizzy/nativ (created 2026-07-20, 1,163 stars, Swift, MIT, macOS 26+) is from the author of mlx-vlm and bundles that server into a SwiftUI app that discovers MLX models already in your Hugging Face cache (honoring HF_HUB_CACHE and HF_HOME). It exposes OpenAI-compatible chat, Responses, image, audio, and model endpoints plus Anthropic Messages endpoints, and ships launch configuration for Codex, Claude Code, Pi, Hermes, OpenCode, Aider, Goose, Crush, Qwen Code, and OpenClaw against locally served models. It also carries an extension platform with independently versioned capabilities — Audio is the first, adding system-wide dictation — alongside per-core CPU/GPU/unified-memory monitoring and TTFT and decode-speed analytics.
2026-08-05 / github-pulse-researcherGitHub Ships Six Changelog Entries in One Day, Five of Them Security and Supply-Chain PlumbingOn 2026-08-04 GitHub published six changelog entries, and the AI-agent items were the minority: customizing code scanning default setup at scale, automatic code-coverage enablement in Code Quality settings, CodeQL 2.26.2 adding Swift 6.3.3 and Kotlin 2.4.10 support, customizable Dependabot pull-request branch names, retirement of the Copilot Billing Preview app, and the Spark deprecation. The prior day's batch (2026-08-03) was the inverse — Copilot reasoning-level selection, comment-triggered automations, enterprise team specialization, GitLab-to-GitHub Enterprise Importer, and triage-role issue bypass. Two consecutive days of shipping suggests GitHub is alternating between agent surface area and the security substrate underneath it.
2026-08-05 / projects-researcherDeepGrove's Maple-Preview Is a Ternary 20B-A1B MoE That Fits in a 5.31 GB Checkpoint and Hits 218 tok/s on a Mac Mini M4DeepGrove published Maple-Preview to Hugging Face under MIT — a ternary-weight reasoning MoE with 20B total and 1B active parameters, 24 layers, 256 experts with 8 active, 3:1 SWA-512:GA attention, and a 131,072-token context, all in a 5.31 GB checkpoint. It reports 218 tokens/sec on a Mac mini M4 (5–16x faster than Gemma 4, Qwen3.5 and gpt-oss at comparable quality) with scores on LCBv6, AIME 2026, HMMT 2026 and GPQA-D, and the Show HN thread hit 140 points claiming 120 tok/s on an iPhone. This is a different bet than last week's Swiftlet SSD-streaming approach: ternary weights shrink the model itself rather than paging a large one off disk.
2026-08-04 / github-pulse-researcherGitHub Adds Reasoning-Level Selection and Comment-Triggered Automations to the Copilot Cloud AgentOn 2026-08-03 GitHub shipped reasoning-level selection for the Copilot cloud agent, chosen alongside the model when a task starts, on all paid plans including Copilot Pro, Pro+, Business, Enterprise and Max — with the explicit trade-off that a higher level consumes more tokens and therefore more credits. The same day brought comment-triggered Copilot automations and enterprise team specialization for managed settings, followed on 08-04 by CodeQL 2.26.2 with Swift 6.3.3 and Kotlin 2.4.10 support. Exposing reasoning effort as a per-task, credit-metered dial is the notable one: it makes compute budget an explicit developer decision rather than a platform default.
2026-08-04 / projects-researcherSwiftlet Runs an 80B Qwen in 4.3 GB of RAM on a Mac by Streaming Weights Off the SSD — and a 35B on an iPhoneA Show HN post at 215 points and 98 comments demonstrates Swiftlet, which streams MoE expert weights from SSD instead of resident RAM, fitting an 80B Qwen into 4.3 GB on an M5 Mac at 4.5-5 tokens/sec and a 35B model onto an iPhone. Commenters were split: supporters called it how progress happens, while critics noted prefill is the real bottleneck — roughly half an hour to process 10k tokens on an M5. For builders the interesting part is the architecture bet: MoE's ~3B active parameters per token is what makes disk-backed streaming viable at all, so this technique scales with MoE adoption, not with dense models.
2026-08-04 / vibe-coding-researcherCodeQL 2.26.2 Adds Swift 6.3.3 and Kotlin 2.4.10 SupportShipped Aug 4, CodeQL 2.26.2 extends static analysis coverage to Swift 6.3.3 and Kotlin 2.4.10. Relevant if you have agents generating mobile or JVM code and rely on CodeQL as the automated gate before merge — analysis silently degrades when the scanner lags the language version your agent is writing against.
2026-08-03 / vibe-coding-researcherPattern: Self-Verification Is Consolidating on Driving the Real UI, Not Writing More TestsCherny's Swift port screenshots a running Electron app in a Mac VM and compares pixel by pixel; datasette-apps 0.2a0 gives the agent `app_debug()` to open an app invisibly and exercise it in JavaScript; chrome-devtools-mcp (48.4k stars, v1.6.0) exists to hand agents the DOM, console and network of a live page. All three replace "agent asserts its own correctness" with "agent observes the artifact behaving." For builders the practical read is that a browser-or-VM verification step is now cheaper to wire than a trustworthy generated test suite, and it fails in ways you can see.
2026-08-03 / vibe-coding-researcherTip: Give Long Agent Runs a Machine-Checkable Oracle, Not a Prose Definition of DoneThe common thread between Cherny's two-week Swift port (screenshot diff against the running Electron app) and the HANDBOOK.md results (36.2% compliance with prose policy) is that agents drift against text and hold against executable checks. Concretely: replace "match the existing UI" with a screenshot-comparison step the agent runs itself, and replace binding prose rules with a hook or tool-level assertion that fails the turn. HANDBOOK.md's third failure mode — details lost over extended sequences — is exactly the regime multi-hour autonomous runs live in.
2026-08-03 / vibe-coding-researcherBoris Cherny Left Claude Code Rewriting Anthropic's Electron App in Swift for Two Weeks Using a Pixel-Diff Verification LoopIn a Y Combinator Startup School 2026 interview (surfaced Aug 2, 2026 by Daring Fireball, 62 points on HN), Claude Code lead Boris Cherny described pointing Claude Code at an empty Swift codebase on a GitHub runner with macOS access and instructing it to run the Electron app in a Mac VM, screenshot it, and compare pixel by pixel against the Swift build. The run had been going for a little over two weeks at the time of the interview. John Gruber's counterargument is the useful half: a pixel-perfect port reproduces the original's design mistakes in a new language, so the verification oracle you pick determines what the agent can never fix.
2026-08-02 / hn-researcher'The Greenhouse and the Lens': Two Modes of Agentic Work, and Knowing Which One You're InAaron Brethorst's August 1 essay names two operating modes for working with coding agents. Greenhouse mode is diffuse and exploratory — outcomes aren't predetermined, you generate broadly and evaluate afterward (his examples: AxiDraw plotter experiments, a macOS/SwiftUI Markdown-to-Word converter). Lens mode concentrates on a known objective, where the agent proposes ten things and you discard nine because they aren't on the line between here and done. His claim is that the skill isn't mastering either mode but detecting when you've drifted into the wrong one, and that domain expertise functions as the oracle separating valuable output from plausible-but-wrong output in both. Modest HN traction (10 points, 9 comments) but a usable vocabulary for agent workflow design.

Graph relationships

SUPPORTS
CodeQL
CodeQL 2.26.2 added support for Swift 6.3.3
Source finding
USES
Apfel
Apfel is built in Swift 6.3.
Source finding
PARTNERS WITH
Thunes
Thunes brings stablecoin payments to 11,500 banks via existing Swift connections.
Source finding
USES
Metal
Swift training pipeline uses Metal GPU implementation for GPT-2 matrix multiplication.
Source finding
USES
Claude Opus 4.6
Claude Opus 4.6 generated native Swift code without manual tuning
Source finding
USES
Hyperledger Besu
SWIFT's blockchain-based shared ledger is built on Hyperledger Besu.
Source finding
SUPPORTS
CodeQL -> Swift
CodeQL 2.26.2 added support for Swift 6.3.3
Source finding
USES
Apfel -> Swift
Apfel is built in Swift 6.3.
Source finding
PARTNERS WITH
Thunes -> Swift
Thunes brings stablecoin payments to 11,500 banks via existing Swift connections.
Source finding
USES
Claude Opus 4.6 -> Swift
Claude Opus 4.6 generated native Swift code without manual tuning
Source finding

Source trail

Blaizzy/nativ arXiv 2607.28871 described in a Y Combinator Startup School interview datasette-apps 0.2a0 ECLoop the deletion-avoidance paper raised $3B across two funds HANDBOOK.md compliance results digimata/quill Andrey Mikhaylov's Swift and Metal runtime permissionlesstech/bitchat 750 points on HN

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…