Skip to content

Run frontier MoE models on hardware you already own β€” pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

C

docs(readme): show that one command line runs any of the three models

JustVugg3937c00770e5
diff --git a/README.md b/README.md
index 713a8fe8..ff4a7bca 100644
--- a/README.md
+++ b/README.md
@@ -361,6 +361,41 @@ On Windows the same commands work with `python coli chat --model D:\glm52_i4`.
The engine at runtime is pure C β€” python is only used by the one-time converter
and the optional API gateway.
+#### The same commands run any of the models
+
+`coli` reads the model's `config.json`, picks the matching engine binary, and
+renders that family's chat template β€” so **nothing about the command line
+changes between models**. Build the engine you want once, then just point
+`COLI_MODEL` at the right directory:
+
+```bash
+make -C c glm # GLM-5.2
+make -C c inkling # Inkling
+make -C c kimi_k3 # Kimi K3
+
+COLI_MODEL=/nvme/glm52_i4 ./coli chat # TUI
+COLI_MODEL=/nvme/inkling_i4 ./coli chat
+COLI_MODEL=/nvme/kimi_k3 ./coli chat
+
+./coli web --model /nvme/inkling_i4 # API + dashboard, same port
+./coli web --model /nvme/kimi_k3
+./coli serve --model /nvme/inkling_i4 # API only
+```
+
+For the non-GLM engines `coli chat` starts the gateway locally and attaches the
+TUI to it, so the TUI, the API and the dashboard all go through the same
+arch-aware chat template β€” you never have to pass the template yourself.
+
+Two things that differ per model, both documented in the per-model page:
+
+- **Inkling on a RAM-tight host** needs the int4 dense container and a small
+ expert cache: `./coli chat --model /nvme/inkling_i4 --cap 2`
+ (see [inkling.md](docs/inkling.md) β€” the default `--cap 8` wants ~14 GB of
+ cache on top of the resident set).
+- **Kimi K3** streams its MXFP4 experts from the original checkpoint, so there
+ is nothing to convert β€” but the snapshot is ~1.6 TB
+ (see [kimi_k3.md](docs/kimi_k3.md)).
+
### 4. Go deeper
| topic | doc |