Ollama-C-lient

C program for interacting with Ollama server from a Linux terminal. It is not meant to be a complete library at all. At the moment, it's just the simplest interface that I'm developing solely for my personal and daily usage.

Btw, because of this, the development of ChatGP-Terminal it is not more maintained by now.

Version:

• 0.0.1-beta (experimental)

Note: since Im moving on agents developing and OS integration, the View layer (Ollama-C-lient.c) will be deprecated soon.

Features/Assumptions/Scope/Whatever

• it supports window/memory context (dynamic and statically)
• it supports 'pipe' redirection
• it supports history of queries during the session
• it supports changing system-roles, models, and adding pre-defined instructions to prompt
• at the moment, the output is only streamed (except when piped)
• allows to show the response's info
• allows to hide/show the model's 'thoughts' (like DeepSeek-R1 and latest models)
• supports only SSL/TLS (sorry, security by design, baby)

Dependencies:

• libssl.so.3
• libcrypto.so.3
• libreadline.so.8
• libc.so.6

sudo apt-get install libssl-dev libcrypt-dev libreadline-dev

Program compilation:

git clone https://github.com/lucho-a/Ollama-C-lient.git

cd Ollama-C-lient/src/

gcc -o ollama-c-lient Ollama-C-lient.c lib/* -lssl -lcrypto -lreadline

Usage:

• $ ollama-c-lient [options]

The options supported are:

Parameter	dataType:defaultValue [boundaries]	Description
--version	N/A:N/A
--server-addr	string:127.0.0.1	URL or IP of the server
--server-port	int:443 [1-65535]	Listening port. Must be SSL/TLS.
--model-file	string:NULL	File with the default model and parameters regarding to context, tokens, etc.
--setting-file	string:NULL	File with the program settings: server info., timeouts, fonts, etc.
--roles-file	string:NULL	File with the different roles for the model.
--instructions-file	string:NULL	File with the different instructions for the model.
--context-file	string:NULL	File where the interactions (except the queries ended with ';') will be stored.
--static-context-file	string:NULL	File where the interactions included into it (separated by '\t') will be include (statically) as interactions in every query sent to the server. This interactions cannot be flushed, and they don't count as 'MAX_MSG_CTX' (it does as 'MAX_TOKENS_CTX').
--show-response-info	N/A:false	Option for showing the responses' information, as tokens count, duration, etc.
--show-thoughts	N/A:false	Option for showing what the model is 'thinking' in reasoning models like 'deepseek-r1'
--stdout-parsed	N/A:false	Option for parsing the output when piped (particularly useful for speeching the response)
--stdout-chunked	N/A:false	Option for chunking the output by sentence. Only works if '--stdout-parsed' was set.

Note: all options are optional (really?!).

On the other hand, some commands can be prompting:

Command	Argument	Description
flush;	N/A	the context in RAM will be cleared (this action won't delete any messages in the context file). I find it useful for avoiding any "misunderstanding" when I start or changing to a new topic.
models;	N/A	shows available models.
model;	string*	change the model.
roles;	N/A	show available roles;
role;	string*	change the system role (included into the file '--roles-file').
instructions;	N/A	show available instructions;
instruction;	string*	add the instructions entered (included into the file '--instructions-file') to prompt history.

• If 'string' is empty, the model|role will change to the default on (included into the model file, if it was specified).

Considerations

• The sent messages are kept into the prompt history, so you can retrieve and editing them using the key up & down arrows.
• The sent messages & the responses are written into the context file if '--context-file' is specified.
• If '--context-file' is specified, the last '[MAX_MSG_CTX]' (parameter in modelfile) messages/responses are read when the program starts as context.
• So, if '[MAX_MSG_CTX]' > 0, and '--context-file' is not set up, the program will start without any context. Nevertheless, as long as chats succeed, they will be stored in RAM and taken into account in the successive interactions.
• If '[MAX_MSG_CTX]' = 0, the interactions won't be recorded into context file.
• The template of modelfile that must be used with '--model-file': here.
• '--setting-file' allows set different parameters. An example of file that should be used: here. The parameters set up in this file, override any parameter passed with any other option.
• the font format in 'settings' file must be ANSI.
• if not setting file is specified, uncolored is set by default.
• if '--roles-file' is specified, prompting 'role;' + the name of the role (specified into the file. V.gr.: role;newRole), will include the (new) role description as system role in the following chats. Example of file: here
• if '--instructions-file' is specified, prompting 'instruction;' + the name of the instruction (specified into the file. V.gr.: instruction;newInstruction), will include the (new) instruction description into the prompt history so you can selected upping/downing the arrow keys. Example of file: here
• If the entered prompt finish with ';', the query/response won't take into account the current context ([MAX_MSG_CTX]), won't be written to the context file, and won't be part of subsequent context messages.
• In case that you want to input a new line without submitting, just use 'alt+enter'. Same key combination for exiting when empty prompt.
• In case that pipes are used, like: 'df | ollama-c-lient >> file.txt', the output is always uncolored, no-streamed, and in RAW format.
• Crl-C cancel the responses.

Examples:

Suggested:

- Chatting

$ ollama-c-lient --model-file ~/ollama/mymodelFile --setting-file ~/ollama/settingFile --static-context-file ~/ollama/staticContextFile --context-file ~/ollama/contextFile

- Scripting/Agents

$ ollama-c-lient --server-addr serverIP --server-port serverPort --model-file ~/ollama/mymodelFile --static-context-file ~/ollama/staticContextFile --stdout-parsed --stdout-chunked --context-file ~/ollama/contextFile

... with a 'model-file' like:

[MODEL]
modelNameAlreadyPulled
[KEEP_ALIVE]
30
[TEMP]
0.6
[MAX_MSG_CTX]
5
[MAX_TOKENS_CTX]
8196
[SYSTEM_ROLE]
You are the assistance and
bla,
bla, bla...

Note: since the incorporation of reasoning models, is not recommended incorporating a 'system-role'. Instead, just leave it blank and incorporate the instructions as part of the 'user-role' (see 'instructions;' and 'instruction;').

Others

$ ollama-c-lient --server-addr 192.168.2.10
$ ollama-c-lient --setting-file ~/ollama/settingFile --instructions-file ~/ollama/instructionsFile
$ ollama-c-lient --server-addr 192.168.1.50 --model-file ~/ollama/any1modelModelFile --context-file ~/ollama/context --show-response-info
$ ollama-c-lient --server-addr myownai.com --server-port 4433 --model-file ~/ollama/any2modelModelFile
$ ollama-c-lient --model-file ~/ollama/any3modelModelFile --show-thoughts
$ ollama-c-lient --server-addr 192.168.2.10 --server-port 4433 --roles-file ~/ollama/roles --context-file ~/ollama/context --static-context-file ~/ollama/staticContextFile
$ (echo 'What can you tell me about my storage: ' && df) | ollama-c-lient --server-addr 192.168.5.123 --server-port 4433 --model-file ~/agents/modelFile --context-file ~/agents/dfAgentContextFile --stdout-parsed >> log-file.log
$ (echo 'What can you tell me about current processes: ' && ps ax) | ollama-c-lient --server-addr 192.168.5.123 --server-port 4433 --model-file ~/agents/modelFile --context-file ~/agents/dfAgentContextFile --stdout-parsed --stdout-chunked | grep 'Ollama-C-lient'

Bugs known/unknown

In line with a Top Down approach (and the beta & experimental stage), bugs will/can be found and the scoping and approach can change at any time.