[OMEGA-427] Use tools API to pass the list of tools to the LLM by vsbogd · Pull Request #358 · singnet/Omega

vsbogd · 2026-09-21T14:54:24Z

Description

This PR is part of #349 which doesn't include loop and memory changes. It delivers the most massive code change: using tools API to get the list of tool calls from LLM. It obsoletes helper.balance_parenthesis function.

How Has This Been Tested?

This PR doesn't introduce any code logic change just migrates existing logic to the new API. Regression testing should be enough. This PR is checked using automatic regression testing and manual smoke check with each provider.

Checklist

PR contains autogenerated code
Self-review completed
Test scenarios above are passed with the version of the code from PR

Move provider specific logic into prepare_args method mostly. The only exceptions are OpenAI because it uses completely different method to call the API and TestMock which doesn't require most of the things.

Change the LLM call API to pass the list of tools to call and receive the list of tool calls. Adapt loop.metta to this change. Fix unit tests.

The historical information is still returned via HISTORY paragraph of the system prompt.

paul-v-snet

Other non-critical problems:

&lastresults and LAST_SKILL_USE_RESULTS have been removed from loop.metta, but are still described in the docs/, e.g., here: https://github.com/vsbogd/Omega/blob/125889524b707e1ee55d12df3fa7bcf9f67c2a67/docs/reference-internals-loop.md;

Also, most of the issues found during QA in #349 still apply to this PR, so they should be addressed here as well.

paul-v-snet · 2026-09-25T10:49:53Z


-def _log_raw(provider: str, model: str, raw: str) -> None:
-    logger.debug(f"[LLM_RAW] provider={provider} model={model} chars={len(raw or '')} raw={raw!r}")
+def _log_raw(kind, provider: str, model: str, raw: Dict) -> None:


non-critical: kind without annotation

suggesting to replace it with: kind: str

paul-v-snet · 2026-09-25T10:53:55Z

-                    raw = llm._llm_empty_response_command()
-            return self._clean_text(raw)
+            if incomplete_reason == "max_output_tokens":
+                response.add_tool_call(_llm_empty_response_tool_call(raw.id))


critical: _llm_empty_response_tool_call defined in providers/lib_llm_ext.py, so it will lead to an error.

suggesting to replace it with: response.add_tool_call(llm._llm_empty_response_tool_call(raw.id))

paul-v-snet · 2026-09-25T10:55:00Z

+        }
+
+    def convert_response(self, raw):
+        response =  LLMResponse()


non-critical: double spaces

paul-v-snet · 2026-09-25T11:04:15Z

+(= (llmRequestMessage role $role content $content)
+   (py-call (providers.llmRequestMessage $role $content)))
+
+(= (llmRequestMessage role $role callid $callid content $content)
+   (py-call (providers.llmToolCallResponseMessage $role $callid $content)))
+
+(= (llmRequestMessage role $role calls $calls content $content)
+   (py-call (providers.llmToolCallMessage $role $calls $content)))


non-critical

I'm afraid this may be confusing. The same expression llmRequestMessage refers to several different things at the same time:

User's request to the LLM

LLM's response

Tool call results

So I believe it would be better to use different names for them, but since this is not critical, it would be enough to just take this into account for the future.

paul-v-snet · 2026-09-25T11:11:21Z

+        self.max_tokens = 6000
+        self.reasoning_mode = "medium"


non-critical, but important

This may lead to hidden issues in the future since max_tokens and reasoning_mode are hardcoded here. If for any reason these values are not passed explicitly, they will be silently configured using these hardcoded values rather than the values from config.yaml.

So I suggest using config_get_by_key here instead of any hardcoded values.

paul-v-snet · 2026-09-25T11:23:52Z

+def _validate_response(request: LLMRequest, response: LLMResponse) -> LLMResponse:
+    for call in response.calls:
+        if call.is_error():
+            continue
+        if not request.has_tool(call.name):
+            call.set_error(f"Unknown tool: {call.name!r}")
+        call.set_tool(request.get_tool(call.name))
+        for parameter in call.tool.parameters:
+            if not parameter.name in call.arguments:
+                call.set_error(f"Call tool parameter is not set: tool: {call.name!r}, parameter: {parameter.name!r}")
+                break
+    return response


potentially critical:

There is no continue after call.set_error(f"Unknown tool: {call.name!r}"), so call.set_tool(request.get_tool(call.name)) at line 248 will return None, and for parameter in call.tool.parameters: at line 249 will raise an AttributeError.

As a side effect, the LLM will never receive the Unknown tool message and will not be able to correct itself in subsequent responses.

paul-v-snet · 2026-09-25T11:27:06Z

+def llmToolCallToSExpr(call: LLMToolCall):
+    sexpr = f"({call.name} "
+    for parameter in call.tool.parameters:
+        if parameter.name in call.arguments:
+            arg = call.arguments[parameter.name]
+            arg = arg.replace('"','\\"')
+            sexpr = sexpr + f"\"{arg}\" "
+    sexpr = sexpr + ")"
+    if call.is_error():
+        sexpr = f"(Error {sexpr} \"{call.error}\")"
+    return f"({call.id} {sexpr})"


Same as what @TossSky found during QA in #349, point 8:

llmToolCallToSExpr escapes " but not \ (providers.py:316), so ls C:\ becomes (shell "ls C:") and the backslash escapes the closing quote. A literal \n in an argument turns into a newline. A non-string argument raises AttributeError at loop.metta:79, outside the catch on line 83.

paul-v-snet · 2026-09-25T11:39:48Z

                             "\nResults in " (researchDir) "/" $research-name "/"
                             "\nWorkflow unloaded from context.")))
-              (appendToHistory  ((get_time_as_string) (newline) (strings-concat  ("RESEARCH [" $research-name "] COMPLETED")) (newline)) )
+              (addToHistory  (strings-concat  ("RESEARCH [" $research-name "] COMPLETED")))


potentially critical:

addToHistory expects 3 arguments, but the workflow plugins provide only 1 everywhere.

I guess this will lead to errors or these expressions being silently ignored (I'm not sure how this will be processed by PeTTa).

Anyway, I'd like to ask @TossSky to check this during QA later.

vsbogd added 11 commits September 21, 2026 15:40

Add embedded skills via common API call add-skill

3bfa8e1

Remove custom OpenRouter client initialization

668c23c

Remove code duplication

ffffa15

Move provider specific logic into prepare_args method mostly. The only exceptions are OpenAI because it uses completely different method to call the API and TestMock which doesn't require most of the things.

Use API fields to pass list of tools to the LLM

dde1860

Change the LLM call API to pass the list of tools to call and receive the list of tool calls. Adapt loop.metta to this change. Fix unit tests.

Remove spamShield flag

ac44ea6

Remove balance_parentheses() function and its dependencies

7afa1d5

Merge remote-tracking branch 'asi/main' into use-tools-api

ec840dd

Merge remote-tracking branch 'asi/main' into use-tools-api

d926b62

Fix logging in OpenAI provider implementation

c795992

Don't send empty input where no user message is sent

1ffb101

Return results of the previous calls via OpenAI API fields

1258895

The historical information is still returned via HISTORY paragraph of the system prompt.

vsbogd marked this pull request as ready for review September 23, 2026 09:03

vsbogd requested review from blackhammer116, jazzbox35 and paul-v-snet September 24, 2026 10:56

vsbogd changed the title ~~[OMEGA-374] Use tools API to pass the list of tools to the LLM~~ [OMEGA-427] Use tools API to pass the list of tools to the LLM Sep 24, 2026

This was referenced Sep 24, 2026

[OMEGA-374] Omega V2: using tools API fields to pass the list of tools to LLM #349

Open

[OMEGA-429] Update channel mock suites for the V2 mock #363

Open

paul-v-snet requested changes Sep 25, 2026

View reviewed changes

Navigation Menu

[OMEGA-427] Use tools API to pass the list of tools to the LLM - #358

[OMEGA-427] Use tools API to pass the list of tools to the LLM#358
vsbogd wants to merge 11 commits into
singnet:mainfrom
vsbogd:use-tools-api

vsbogd commented Sep 21, 2026 •

edited

Loading

Uh oh!

paul-v-snet left a comment

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

paul-v-snet Sep 25, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

vsbogd commented Sep 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

How Has This Been Tested?

Checklist

Uh oh!

paul-v-snet left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

vsbogd commented Sep 21, 2026 •

edited

Loading