built a small CLI tool that moves your AI coding session context between Claude Code, Gemini, Copilot, Codex and OpenCode when you hit rate limits. no more re-explaining what you were doing, or painful copy/paste from terminal that lose tool-call context. just `npx continues`

show hn today: repo:

Pretty funny how the world builds the same shit at the same time... I've got a nononesense version of this. written in sqllite. I bounce between Claude, Codex, Droid (GLM 5|KimiK2.5|MiniMax2.5) and Kilo-open-Code (to run copilot sub) Basic af, but works - AGENTS md: "After every session - handoff into sqlite."

sqlite is great idea! copilot sdk uses same pattern on their fleet agent system and better than handling jq over jsonl

first new feature: @droid support thx @DCasota

It looks pretty good, Yigit. I will give a shot. ✨

it'd be an honor to build a utility that even an elite typescript dev like you would actually use 🫡

wow, 200K+ impressions. that got me excited to share more. so here's the challenge: sharing a new repo from my own workflows every day for the next 30 days @steipete's github profile readme before building @openclaw was the inspiration next one [2/30]:

@openclaw CLI party continues [3/30] - this one helps you to kill an app by referencing port. was expecting this to be in part of unix's default coretools but somehow, it is not. so i've built an easy-to-use version:

amazing work, gotta try this one

i should print this tweet out and tape it to the edge of my desk lol probably 90% of my system architecture know-how (which helped me A LOT) came from your YT videos during covid + books you read on stream. if it’s my turn to give something back, that would make me happy 🥂

just tried now and I think it doesn't quite work the way I expect. the handoff file basically only has the last prompt from CC. is it not possible to dig up latest prompts from Claude session file or history file?

i can implement a recursive session lookup, basically grab the prev session id when we spot one and stitch everything together like it was one long thread. normal CC behavior just spins up a clean session, but if it's a continuation the first msg is usually a compact summary. i dig the idea, lemme code it real quick and ping you back.

Yeah it lost all the useful context of what we were doing when I switched from CC to Cursor and handoff file was basically just the last prompt and even that was truncated like (...)

it was planned to compress even more for some large context cases where the new context is too big for the new coding agent. then it has to compress even more, like showing only the first 500 characters of some messages, but it probably triggered before it actually became an issue. will fix everything in a few hours, thx!

useful tool but this is duct tape over a deeper problem. every new coding agent means another migration path to maintain. what we actually need is a shared context protocol, like how LSP standardized language features across editors. without that, continues will need an update every time someone ships a new agent.

Did something similar, with encrypted shares too if you wanted to share externally. In the same folder so just need to tell the other app to read .context and continue

very cool stuff! @yigitkonur

we should integrate such thing to @emdashsh - so that I can migrate between sessions 🤤

@emdashsh Definitely! 🫡

awesome! also, a very clean screen recording. is this CleanShot X?

nothing crazy, just another warp windows vibing in the back. yeah it’s @CleanShot, but the magic’s all in how the theme and frontend play together, @warpdotdev theme + font combo absolutely slaps

@CleanShot @warpdotdev Awesome! Thank you, i have @CleanShot but didn’t know it can have background for recording as well as screenshots. It’s clear and beautiful

Pretty awesome.

This solves a real pain point. I bounce between Claude Code and Codex and the context re-explanation overhead is brutal. Curious how well it preserves structured tool-call state vs just conversation history — that's where transfers usually break down.

this is basically a session-portable harness. the next step is making the context handoff model-aware — each harness has different tool call schemas so naive context transfer loses the tool-call graph. curious if you serialize the full tool history or just the conversation text.

this is a question can be asked by real expert! markdownifying less verbose tool call (or truncated first 100 chars) but planning to add 'full' migration mode that even follows schema of agent tools mapped and converted too.but even with that way it can help 90%+ of regular cases

Super useful tool, love the name too! 👏

with your PR it becomes even better!

@developedbyed

Genius

Super abi eline saglik

this is fucking awesome brother you rock

this solves one of the most annoying parts of AI-assisted coding. losing context when you switch tools is brutal, especially mid-refactor. the fact it preserves tool-call context too is huge

🚀🚀🚀

Less is more dedikleri tam da bu ya, HARİKA!

Automating context transfer across platforms is a game-changer for efficiency.

Next level 👏 Gonna check whats good, perhaps can have some QoL updates to my simple /handoff-context skill 😅 For context, just writes a smol temp YAML on invokation:

very cool. i literally built a similar feature yesterday that handsoff context to other agents when hitting rate limits.

God Job!

oh this is genuinely useful. the worst part of hitting rate limits isn't the wait, it's losing all the context you built up over the last hour. does it preserve tool call history too or just the conversation?

Wow

rate limit musical chairs is the most frustrating part of using multiple AI coding tools right now. you hit the ceiling on Claude Code, switch to Codex, lose all your context, spend 10 minutes re-explaining the codebase, and by the time you are productive again the original rate limit has reset. preserving the full tool-call context across providers is the key insight here because that is what takes the longest to rebuild manually. a dev running a SaaS studio in Riyadh told me he keeps three terminals open with different providers just to avoid downtime and the context switching cost was killing his velocity. npx continues would solve that workflow entirely

Bro i tried making this but gave up, thanks

Rate limit context loss is the most underrated friction point in agentic workflows right now. The missing piece was never better prompts but portable session state that survives a tool switch.

This has always been the end goal, the convergence at the end. I always knew they will merge at the end and become a single overlord. Anyone can tell where it is heading, it’s a raise to the bottom.

awesome, will it support @AmpCode and cursor agent ?

@AmpCode yes, @Evrim267 created PR for cursor agent, will merge soon. for @AmpCode I can build sth and push today

@fakechris @AmpCode I make an another update and pull request. test it different situations. looks good so far.

@fakechris @AmpCode thx, just merged!

Rate limit handoff is the unsexy problem nobody talks about. The context loss mid-session is brutal. Covered how Claude and Codex differ on long-session resilience:

@grok does opencode do the same?

This solves the exact pain I hit every day. Re-explaining context after a rate limit is the worst productivity killer. Bookmarked.