Score breakdown
Popularity is tracked separately. Support, ads, sponsorships, and tips never affect these signals.
Why it matters
Useful for developers who want to run a frontier-scale open-weight model on hardware they already own instead of renting API access; for people evaluating whether expert streaming is fast enough for their workload before buying GPUs; for systems engineers who want a small, readable codebase where an inference optimization can actually be tested.
Who should use it
Who should skip it
Skip colibrì — pure-C inference engine that runs 744B to 2.8T MoE models by treating disk, RAM and VRAM as one memory hierarchy unless the captured evidence suggests it solves a problem you are actively working on.
About this signal
colibrì — pure-C inference engine that runs 744B to 2.8T MoE models by treating disk, RAM and VRAM as one memory hierarchy is tracked by RepoRadar as a developer tool in the Local Inference section. First seen —; the source record was last checked on 2026-08-30. The current verdict is 'try now' with a Gold tier and moderate setup difficulty. The standout signals for colibrì — pure-C inference engine that runs 744B to 2.8T MoE models by treating disk, RAM and VRAM as one memory hierarchy are momentum (10.0) and workflow potential (10.0), while setup ease (6.4) trails — that balance shapes where it fits best. This page summarizes the public evidence on the linked source page and states where additional review is still needed.
How this item is evaluated
The colibrì — pure-C inference engine that runs 744B to 2.8T MoE models by treating disk, RAM and VRAM as one memory hierarchy record combines a 8.9/10 composite score with separate popularity (100.0), risk (none), and setup (moderate) signals. See the scoring methodology for the current weights and evidence definitions.
Questions worth asking before you adopt this
Putting this into practice? Read How to evaluate an AI tool before you adopt it for the checklist behind this score.
Risk explanation
The maintainers state there is deliberately no SLA on speed: this is an inference engine and an open research platform, and placement policies are described as measurable experiments rather than guarantees; Throughput figures in the README are maintainer-measured on specific hardware; the same model on a different SSD or GPU can land far from them. 81 open issues at time of review; Large-model use means large downloads and sustained NVMe reads; plan disk capacity before starting; Overlaps in purpose with sqliteai/warp, already covered on RepoRadar. They are independent projects by different authors; colibrì is the larger and more general one (more model families, GPU backends, web dashboard).