GitHub (andrewboldi/tmux-restore) · Nov 3, 2026 · 6:20 PM

tmux-restore brings back tmux sessions after a reboot and reopens Claude Code, Codex and Gemini CLI conversations

Shown on HN Nov 3 under MIT, the one-line installer wires tmux-resurrect and tmux-continuum to save sessions, layouts and scrollback every minute and once more at shutdown, while small adapters find each agent's conversation id so Claude Code, Codex, Gemini CLI, opencode and pi panes resume mid-conversation. Editors restart on their own, but deploys and dev servers come back typed on the prompt, waiting for Enter.

Read at github.com

Claude Code

An AI coding agent for terminal, IDE, web, and Slack

More news

View all

DrawDB can now turn a live database into an ER diagram from a single query

Shown on HN Nov 3, the free, no account browser ERD editor adds File Import from database in its editor, mapping an existing schema into a diagram you can then edit, export as SQL or turn into migrations. It's open source and self hostable with Docker; sharing needs the separate drawdb server (Show HN)…

GitHub (drawdb-io/drawdb) 27 minutes ago

Duckity swaps CAPTCHAs for invisible rate limits and proof-of-work challenges to make API abuse costly

Shown on HN, await duckity.solve(POLICY ID) has the client do a VDF backed computational challenge that scales with activity, alongside rate limits and IP signals, so bots and scripted agents pay more to hammer your endpoints while users never see a puzzle; data is processed in Frankfurt. Plans list a 10,000 challenge monthly tier with a "Protected by Duckity" label and $49/mo for 100,000 with white labeling…

Duckity (Show HN) 2 hours ago

Toolang v0.4 writes agent loops as functions and flows in plain-text .too files you can run from the CLI

Shown on HN, uv tool install toolang adds a runtime where an agic is a reusable agent loop with typed inputs and outputs, flow composes them in sequence, parallel lanes or loops, and skills, psyches and MCP services plug in with with lines; any .too file's agics become CLI commands with model , allow tools=none and token limits. It works with models.dev cloud providers plus Ollama and llama.cpp, needs Python 3.11+, and is untested on Windows…

Toolang (Show HN) 3 hours ago

Agent Trace lets Claude Code and Codex read each other's local chat histories over MCP

Shown on HN Nov 3 under Apache 2.0, it turns local Codex and Claude Code sessions into compact or detailed context (prompts, thinking summaries, command output, diffs, subagents) that either agent can query through a local MCP server. The author's clipboard app for pasting threads into web chats isn't packaged yet (Show HN)…

GitHub (wcook04/agent-trace-mechanism) 5 hours ago

Hardcount is an MIT, self-hosted usage billing server that prices AI tokens per model and shows margin per customer

Shown on HN Nov 3, the Go server takes usage events (or LiteLLM's) priced by model and token type, sells prepaid credit, makes monthly invoices that Stripe charges with no Stripe Billing fee, and keeps data in your own Postgres, with test clocks to preview a month's invoice. It's pre 1.0 with one maintainer, Stripe and cards only, and no production users yet (Show HN)…

GitHub (gethardcount/hardcount) 6 hours ago

Spotted something we missed? Start a thread.