From Dev.to Community
Tech Articles
Curated developer articles, tutorials, and guides — auto-updated hourly
Mr Say NothingSep 22, 2026 • 4 min read
128K Context on a Desktop Is a Lie. The KV Cache Ate It.
Model cards brag 128k context. The KV-cache arithmetic says your RAM pays up front — 16 GiB for an 8...
#discuss#ai#localai#memory
0 0
Mihai PerdumSep 19, 2026 • 13 min read
goose compaction: my 27B re-read the file 95% of the time
Compaction summarised away the tool output holding the file. Quoting the file into the summary left ...
#aicoding#gooselocaledition#localai
0 0
MichaelSep 23, 2026 • 4 min read
How to Run Local LLMs on Apple Silicon
This is about the model, on its own — what tool to install, what format to pull, and how to actually...
#ai#llm#apple#localai
0 0
MichaelSep 23, 2026 • 4 min read
LLM Quantization Explained for Mac Users
The RAM math post treats quantization as an input — "4-bit is roughly 0.5 bytes per parameter" — and...
#localai#llm#ai#macos
0 0
Mihai PerdumSep 18, 2026 • 6 min read
lsof: command not found in a packaged Mac app, and the port reclaim that never ran
macOS keeps lsof in /usr/sbin, which a packaged app's PATH omits. The spawn ENOENTs, the reclaim nev...
#aicoding#gooselocaledition#localai
0 0
jamilxtSep 18, 2026 • 8 min read
Bonsai 2 27B Puts a 27B AI Model in 5.9GB - Can It Replace Your Paid Subscription?
A strange thing happened to local AI models this year. Every few months, someone released a "2-bit"....
#ai#localai#llm#machinelearning
4 2
Mihai PerdumSep 18, 2026 • 6 min read
LeanZero at Atlassian Team '26 Europe Amsterdam: no booth, here to talk Forge migrations
LeanZero is heading to Atlassian Team '26 Europe in Amsterdam without a booth or speaking slot - fin...
#forge#localai
0 0
Mihai PerdumSep 19, 2026 • 15 min read
goose swarm: pytest | head -80 exits 0 when nothing ran, and pipefail only trades the lie
pytest exits 4 for a missing file; piped to head it exits 0 in bash and zsh. Why a goose swarm worke...
#aicoding#gooselocaledition#localai
0 0
Mihai PerdumSep 19, 2026 • 19 min read
goose Local Edition: a benchmark that runs the app your agent built
Our goose fork, its swarm engine, and a scorer that boots the built app and prices what a user would
#localai#gooselocaledition
0 0
Mihai PerdumSep 19, 2026 • 30 min read
goose swarm benchmark: a scorer that can't flatter itself
How we built an execution-graded benchmark for coding agents, and the six disciplines that caught ev...
#localai#gooselocaledition
0 0
Mihai PerdumSep 19, 2026 • 15 min read
What 145 MCP Tools Cost Before the Model Reads Your Question
Measured on ten real MCP servers: 145 tool definitions burn 40,784 context tokens before your questi...
#aicoding#localai
0 0
Mihai PerdumSep 20, 2026 • 17 min read
Inside goose-swarm: How We Turned One Local Model Into a Self-Verifying Fleet
How goose local edition works: the scheduler, the parallel planner, the CONTRACTS discipline, the mo...
#ai#goose#localai#architecture
0 0
Mihai PerdumSep 23, 2026 • 22 min read
Goose Swarm 3.0: local models on every Mac, joined by LeanZero Link
Goose Swarm 3.0 runs local models on its own MLX engine, joins your Macs with LeanZero Link and serv...
#goose#localai#mlx#ai
1 0
Mihai PerdumSep 20, 2026 • 12 min read
swarm-gym for goose: grading local models by running the code they write
A hands-on swarm-gym tutorial: set up the goose local-edition swarm testbed, run both modes, read ev...
#ai#goose#localai#swarm
0 0
Mihai PerdumSep 20, 2026 • 15 min read
The goose swarm, self-verifying: small local models shipping real software
How we forked Block's goose into a multi-agent swarm running three small local models on LM Studio, ...
#ai#goose#localai#swarm
0 0
Mihai PerdumSep 19, 2026 • 15 min read
MLX memory ceiling on a 96 GB M3 Ultra
Metal reports 77.76 GiB on a 96 GB M3 Ultra, not the 72 GiB the 75% rule predicts. Three ceilings me...
#aicoding#localai
0 0
Mihai PerdumSep 20, 2026 • 25 min read
Qwen3-Coder-Next + MXFP8: The 128GB Local LLM That Runs Predictably
Qwen3-Coder-Next MXFP8: 80B Local on 128GB
#aicoding#localai#llm#mlx
0 0
Mihai PerdumSep 20, 2026 • 7 min read
MLX vs GGUF on Apple Silicon: Benchmarking the Same Local Model Two Ways
The same local model as GGUF and as MLX, graded by swarm-gym running the code it writes. MLX is fast...
#ai#mlx#benchmark#localai
0 0
Mihai PerdumSep 20, 2026 • 17 min read
Making the goose swarm predictable: 602 commits, 100 levers, and three bugs I found writing this
602 commits on goose local edition: what changed in the swarm, what the desktop looks like now, and ...
#aicoding#localai#rust#electron
0 0
Mihai PerdumSep 20, 2026 • 7 min read
v0.5 of our Atlassian Qwen3.8-27B: beats v0.4, now with GGUF for llama.cpp and Ollama
Qwen3.8-27B tuned for Forge and Jira: v0.5 beat v0.4 on every gate. GGUF for llama.cpp and Ollama, Q...
#localai#mlx#gguf#llm
0 0
Mihai PerdumSep 20, 2026 • 12 min read
Convert a Fine-Tuned MLX Model to GGUF: Proving Quantization Parity with KL Divergence
Convert a fine-tuned MLX model to GGUF for llama.cpp, then prove quantization parity: bf16 match, cr...
#localai
0 0
Mihai PerdumSep 20, 2026 • 18 min read
llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x
MTP is merged in llama.cpp, still an open PR in mlx-lm. On an M3 Ultra every out-of-the-box MTP sett...
#aicoding#localai
0 0
Mihai PerdumSep 20, 2026 • 14 min read
MXFP8 vs Q8: 10x the weight error, 1% the perplexity
MXFP8 reconstructs Qwen3-Coder weights about 10x worse than 8-bit affine at identical size, then cos...
#aicoding#localai
0 0
Mihai PerdumSep 20, 2026 • 18 min read
Ling 3.0 Flash won't load on a Mac Studio. Qwen3-Coder-Next does, at 73 tok/s.
Ant Group's Ling 3.0 Flash still won't load in mlx-lm on a 96 GB Mac Studio: 385 of 62,237 tensors b...
#aicoding#localai
0 0