After discovering that Opencode is brilliant for full-on online-LLM’s such as Deepseek-v4-pro, I soon found that it doesn’t play well with local Ollama models, namely they stop talking at undefined moments and are generally useless.
However after trying other Agents with no success either, I’m happy to announce that the “Pi Coding Agent” is showing a lot of promise with locally hosted small Ollama models.
So far my best results have been with gemma4:26b which is a recent 17GB image from Ollama.
The response of this LLM is more like a 4GB agent on the current system, AMD Ryzen 5500 CPU with 64GB ram and a single Nvidia RTX3060 GPU.
In other words, it’s very fast, far faster than I expected, which is in line with the general gemma4:26b press releases.
Most importantly it behaves very well with the Pi Coding Agent and initially has four ‘tools’ it can use on your system to find, write and read files etc.
This is still early days for me in my exploration of Pi, but so far I’m highly impressed.
Is Pi an alternative to Opencode and Deepseek online ? NO, and it never will be while using a local model because they are in
different universes.
My PC: 6 CPU’s and 64GB ram with a single RTX3060 GPU
Deepseek.com: probably has thousands of High end GPU’s, 100’s of TB of video ram TB of Hard drives.
Nevertheless, a local LLM is pretty handy at finding text, doing housekeeping, answering questions, teaching concepts and a zillion other things, so they absolutely have their niche.
If you do try Pi, be sure to check out the massive documentation it installs in /opt/pi-coding-agent/docs.
Here is a small sample of the Pi interaction, being locally hosted the cost is of course: nothing.
can you see the local ollama models installed ? format your output so it renders correctly when pasted into a online Discourse web forum.
To see the locally installed Ollama models, I need to use the ollama list command via bash.
$ ollama list
llama3.1:8b 46e0c10c039e 4.9 GB 10 hours ago
gemma4:26b 5571076f3d70 17 GB 2 days ago
ministral-3:14b 4760c35aeb9d 9.1 GB 2 days ago
mychen76/qwen2.5_cline_roocode:32b a9ceeb5f1467 19 GB 10 months ago
michaelneale/deepseek-r1-goose:70b 90cc3dd53da8 42 GB 10 months ago
Took 0.0s
Okay, I can see the installed Ollama models.
| Name | ID | Size | Modified |
|---|---|---|---|
| llama3.1:8b | 46e0c10c039e | 4.9 GB | 10 hours ago |
| gemma4:26b | 5571076f3d70 | 17 GB | 2 days ago |
| ministral-3:14b | 4760c35aeb9d | 9.1 GB | 2 days ago |
| mychen76/qwen2.5_cline_roocode:32b | a9ceeb5f1467 | 19 GB | 10 months ago |
| michaelneale/deepseek-r1-goose:70b | 90cc3dd53da8 | 42 GB | 10 months ago |