@polytoken.dev
A coding harness by somebody who is kind of obsessive about nice coding harnesses. That somebody is @ed3d.net.
@polytoken.dev PinnedPolytoken stable just advanced to 0.6.13. Lots of bug fixes, some enterprise-friendliness adjustments, new model providers, new models, better compaction, better post-compaction, shell mode with !, bunch of other stuff.
`curl -fsS get.polytoken.dev | bash` is friend.
@polytoken.devI’m working on one in my spare time. It’ll be out when I think it’s good enough to replace uses of the TUI.
@polytoken.devhuh this post did a number, huh
@polytoken.devnobody else has the steel moral core to tell you the truth in release notes
@polytoken.dev(it works. it isn't good)
@polytoken.devoutlook unclear, ask again later
@polytoken.dev
@polytoken.devI have a Windows machine that will be cooking on a port, but in transparency, WSL will always be the standard-rec method because working on Windows makes me feel bad But the GUI app will have automatic WSL integration once I get that weaseled together
@hailey.atReposted by @polytoken.devi basically agree. my multi-model setup right now is openai luna and sol, glm 5.2 (soon to be 5.3!!), kimi k3, and deepseek v4. i removed all anthropic models from my menu.
@polytoken.dev'cause it owns
@ed3d.netReposted by @polytoken.devTool flows are Extremely Experimental and the agent isn't biased towards using them, but I think there are pretty good use cases here, especially when coupled with datastores and other stuff that might dump a lot of text your agent doesn't actually care about.
@polytoken.devPolytoken 0.8.0 released to latest, with a new feature: tool flows. Agents can orchestrate the inputs and outputs of their tool library in a sandboxed environment and return to the harness only what it needs.
Stable has updated to 0.7.4.
@geesawra.industriesReposted by @polytoken.devTree shaking must feel so good
@polytoken.devit could not have had halfer an ass applied
@polytoken.devpolytoken.dev is now a website
plase clap
@ed3d.netReposted by @polytoken.devOn Max, GLM-5.3's token usage is best described as "bullshit" and the high/max performance divide seems very small. Polytoken defaults both 5.2 and 5.3 to high, because I am ungovernable by model provider advice.
@polytoken.devhmm. a good point
@polytoken.dev📣 maaaaany people are saaaaaaaying 📣
@polytoken.devnot yet
@pvg.bsky.socialReposted by @polytoken.devsomeone called a thing deepinfra when derpthra was right there