Issue Nº 01 — MODEL

Hundreds of AI tools came out this week. We kept three. One we put through a fitting ourselves. The other two we could not try on, so they are labeled Not tested and everything about them comes from their makers.

The Cover

Graphicle

A code graph and memory for your coding agent.

Where: github.com/graphicle-ai/graphicle (opens in a new tab) (v0.1.0 released October 14, 2026).

The cut: One local Rust binary. It indexes a repository into a code graph and gives a coding agent — Claude Code, Codex, Cursor, VS Code, or any MCP client — cited, line-anchored excerpts, so the agent stops re-reading the whole repo.

The price: Free and open source. No account, no API key, no telemetry, per the README.

The fitting

Tested, hands-on. We downloaded the v0.1.0 Linux binary, checksum verified, and ran it on a tiny two-file Python repository.

graphicle setup --yes took 0.2 seconds. It listed every file it touched, including AGENTS.md and git hooks, and said no agent client was installed on the machine.
graphicle graph extract . --code-only produced 6 nodes and 10 edges.
graphicle graph affected "price_with_tax()" returned checkout() at shop.py line 5, handle() at api.py line 3, and the api.py import. All correct. That is a real blast-radius check before an agent refactors.
graphicle ask "how are failed requests retried?" returned both files with SHA-256-tagged, cited excerpts inside a 3,000-token budget. It does not write an answer unless a model is configured.

What we did not test: the MCP integration with a live agent, and speed on a large repository. The “8x faster than Graphify on Kubernetes” claim is theirs, not ours. The README’s own timing, one exploratory run, is 3 minutes 18 seconds against 26 minutes 38 seconds for Graphify 0.9.88 on a 31,000-file tree.

The verdict

Try it. The most useful of the three for people whose agents keep getting lost in a growing codebase. Undo it with graphicle disconnect.

Not this season if

You are on native Windows (WSL2 only, untested by them), the project is small enough that the agent already sees everything, or you will not trust a v0.1.0 tool on the main repository. Try it on a side repo first.

The Collection

Rill Browser

A browser where Claude Code and Codex work beside you.

Where: rill.love (opens in a new tab) (launched on Product Hunt in early October 2026).

The cut: A Mac browser. Press Command-E on any page and it becomes a task for your Claude Code or Codex. Rill picks the project from the agent history, and the agent works in background tabs while you keep browsing.

The price: “Rill is free.” AI features run on an existing Claude Code or Codex plan. Beta.

Not tested

It is macOS 14 or later on Apple silicon only, and we had no Mac. Everything here comes from the official site.

How to wear it: You are reading a release note, or looking at a UI you like. Hit Command-E, then ask it to try the idea in a portfolio project, and the coding agent gets the page context with no copy-pasting. Per the site, new tasks are read-only until edits are allowed, and web tasks stop before paying, posting, sending, or deleting.

Rill verdict

Worth a download if you already pay for Claude Code or Codex. It turns “I saw something” into a queued agent task.

Not this season if

You are on Windows, Linux, or an Intel Mac; you do not use Claude Code or Codex; or you are uneasy about an agent browsing with your sign-ins. The site itself says its prompt-injection safeguards lower the risk and do not remove it.

Retriever AI

A free AI browser agent.

Where: rtrvr.ai (opens in a new tab) (on Product Hunt October 14, 2026).

The cut: A browser agent — Chrome extension, cloud browser, API, MCP — for web chores: collecting site data, researching leads, filling forms from documents, prepping job applications.

The price: Free Mode includes supported AI models and cloud browser time with no API key or subscription, paid for by sponsored cards, with a daily fair-use allowance. Schedules, triggers, and Chrome control via API or MCP need a paid plan. We did not confirm paid plan prices.

Not tested

Using it requires an extension and an account, which we did not do for this issue.

How to wear it: Their own task guides include analyzing a TikTok profile into post metrics and a content brief, and researching YouTube competitors. That is useful for a creator sizing up a niche without paying for a scraper.

Retriever verdict

Free, with no API key, makes it the lowest-friction way to see what a browser agent can do. Expect ads.

Not this season if

You hate sponsored content, you need scheduled runs on a free plan, or you do not want an ad-supported agent acting inside logged-in accounts.

The Margins

Not evaluated: Google EmbeddingGemma 2 (open on-device multimodal embeddings, Apache 2.0), CLI-Dispatch (hand Claude Code tasks to Codex, Gemini, or DeepSeek), and Offrun (manage multiple coding agents, on a Mac).

Reply with the tool you want us to test next.