Everlier

@Everlier

Building LLM agents & tools. Local inference will win. github.com/av/harbor Also: Facts / Mi / Skilled / Pace @viktor_com

av.codes

Joined April 2010

Replies Reposts Media Articles

Pinned
Everlier
@Everlier
Mar 20
You don't even need Kimi 2.5 for a decent local LLM setup. - llama.cpp - Unsloth's Qwen 3.5 35B A3B with UD Q4 K XL quants - OpenCode - av/harbor It'll take a while to download/install, but otherwise it's something that mid-range hardware (>32GB RAM, ~8GB VRAM) can run today.
fynn
@fynnso
Mar 19
was messing with the OpenAI base URL in Cursor and caught this accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast so composer 2 is just Kimi K2.5 with RL at least rename the model ID
16
Everlier
@Everlier
9h
GPT-6 Astra quirks and workarounds. Asks too many questions. Fix: "Make reasonable assumptions and complete the task; ask only when missing information materially changes the outcome." Overreacting to skill instructions. "Prioritize my explicit instructions over skill guidance,
Everlier
@Everlier
10h
For GPT-6 Astra, OpenAI explicitly warns that ambiguous or conflicting AGENTS.md and skill instructions can make Astra pause unnecessarily. Cleaning up those files may materially improve autonomy. "Read AGENTS.md, all the skill files and identify all ambiguities and
1
Everlier
@Everlier
10h
GPT-6 Astra can continue reasoning, call other tools, or answer independent parts of a request. E.g. if your harness sets it up - it can launch a long and heavy task while model moves onto other things, eventually being notified when they are done. Think... subagents, maybe
Everlier
@Everlier
10h
GPT‑6 Astra uses 2.5x the credits per token of GPT‑5.6 Sol in standard mode. You'll soon feel that your ChatGPT subscription is not limitless at all.