Show HN: A lightweight way to make agents talk without paying for API usage

Hackers make chatbots gossip for free — DIY tricks, token fears, and terminal drama

TLDR: A hacker shows how to make AI bots talk to each other using your existing subscriptions and a terminal, skipping extra fees. Commenters split between loving the scrappy, fast setup and warning about limits like no parallel chats and surprise costs, while others plug polished tools that do it all for you.

Paywalls, be gone: one Show HN post suggests making AI bots chat using the tools you already pay for—no extra app fees, no complicated setup. The author’s recipe is simple: have one bot “resume” the same conversation and call another bot from the command line, or open split screens with tmux (a windowed terminal) so you can watch them bicker in real time. The crowd? Loud and divided.

Fans of the no-frills approach are cheering. swingboy flexed that they already built a skill for this with OpenCode and GitHub Copilot, saying it “works pretty well.” Others are swapping shortcuts: DeathArrow claims it’s “very easy” in Cursor or OpenCode—just switch the model in the same chat. Meanwhile, the power users are dropping names: scalefirst pointed to roborev.io from data celeb Wes McKinney, and dragonfax brought the big guns with Claude Code Agent Teams, warning these auto-chatty teams can blow through tokens if you forget to stop them.

But the drama hits where it hurts: convenience vs. control. pitched warns that always “resuming” the same session means no parallel runs, which is a buzzkill for folks juggling multiple projects. Others say tmux visibility is a must; watching bots argue in separate panes feels like AI couples therapy—equal parts helpful and chaotic. Bottom line: the community is split between “keep it scrappy and cheap” and “give me pro tools with guardrails,” with plenty of jokes about chatbots gossiping like coworkers around a watercooler.

Key Points

  • The article outlines a method for making coding agents collaborate without APIs by using existing subscription CLIs.
  • Maintaining session continuity is key: invoke agents in resume mode (e.g., codex --last, gemini -r latest) to preserve context.
  • A shared memory file of invocation conventions ensures agents can consistently continue prior interactions.
  • The basic loop: one agent drafts, another critiques via resume mode, and the orchestrator iterates until results stabilize.
  • tmux offers a more visible variant, enabling separate panes/sessions for each agent and easier output capture, at the cost of an extra dependency.

Hottest takes

"Works pretty well." — swingboy
"you can’t have parallel systems going." — pitched
"They can rack up some extra tokens" — dragonfax
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.