feat: expand LLM provider support to include OpenRouter and NVIDIA NIM#67
Open
MauroDruwel wants to merge 7 commits intomemovai:mainfrom
Open
feat: expand LLM provider support to include OpenRouter and NVIDIA NIM#67MauroDruwel wants to merge 7 commits intomemovai:mainfrom
MauroDruwel wants to merge 7 commits intomemovai:mainfrom
Conversation
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughAdds OpenRouter and NVIDIA NIM as supported LLM providers alongside Anthropic and OpenAI; updates docs, config macros, CLI help text, and the LLM proxy to resolve provider-specific hosts/paths/URLs and choose appropriate authorization and request/response formats. Changes
Sequence Diagram(s)sequenceDiagram
participant User as "User"
participant CLI as "CLI"
participant LLMProxy as "LLM Proxy"
participant Provider as "LLM Provider\n(Anthropic / OpenAI / OpenRouter / NVIDIA)"
User->>CLI: submit chat or set provider
CLI->>LLMProxy: forward request + provider selection
LLMProxy->>LLMProxy: resolve URL/host/path/auth (provider_is_openai_compat?)
alt Tool invocation required
LLMProxy->>LLMProxy: build tool call payload (provider-specific fields)
LLMProxy->>Provider: POST (Bearer or x-api-key) to resolved path
else No tools
LLMProxy->>Provider: POST (Bearer or x-api-key) to resolved path
end
Provider-->>LLMProxy: response (messages / tokens / tool outputs)
LLMProxy-->>CLI: normalized response
CLI-->>User: display result
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
📝 Coding Plan
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment Tip You can customize the high-level summary generated by CodeRabbit.Configure the |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
main/mimi_secrets.h.example (1)
20-24:⚠️ Potential issue | 🟠 MajorFix unterminated comment in LLM API section.
The/* ...comment is missing a closing*/, which will comment out the rest of the file if users copy this template.🔧 Proposed fix
-/* LLM API (anthropic, openai, openrouter, or nvidia_nim) +/* LLM API (anthropic, openai, openrouter, or nvidia_nim) */ `#define` MIMI_SECRET_API_KEY "" `#define` MIMI_SECRET_MODEL "" `#define` MIMI_SECRET_MODEL_PROVIDER "anthropic"🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@main/mimi_secrets.h.example` around lines 20 - 24, The LLM API section starts a block comment with "/*" that is never closed, which will comment out subsequent lines including the MIMI_SECRET_API_KEY, MIMI_SECRET_MODEL, and MIMI_SECRET_MODEL_PROVIDER definitions; close the block comment by adding the matching "*/" after the intended comment text (or remove the opening "/*" if the lines are meant to be active) so the constants (MIMI_SECRET_API_KEY, MIMI_SECRET_MODEL, MIMI_SECRET_MODEL_PROVIDER) are not accidentally commented out.main/llm/llm_proxy.c (1)
552-586:⚠️ Potential issue | 🟠 MajorOpenRouter and NVIDIA NIM are OpenAI-compatible but use Anthropic request/response paths.
Request building, response parsing, and tool handling are guarded by
provider_is_openai()only, soopenrouterandnvidia_nimsend Anthropic-shaped payloads and attempt Anthropic-style response parsing despite being OpenAI-compatible endpoints. This breaks both basic requests and tool use.Introduce an OpenAI-compatibility predicate:
static bool provider_is_openai_compat(void) { return provider_is_openai() || provider_is_openrouter() || provider_is_nvidia_nim(); }Then replace
provider_is_openai()withprovider_is_openai_compat()in all request/response handling sections (lines 552–586, 638–642, 681–712, 757–809) for message conversion, token limits, response extraction, and tool call parsing.Also applies to: 638-642, 681-712, 757-809
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@main/llm/llm_proxy.c` around lines 552 - 586, Add a new predicate provider_is_openai_compat() that returns true when provider_is_openai() or provider_is_openrouter() or provider_is_nvidia_nim() is true, then replace calls to provider_is_openai() with provider_is_openai_compat() in all request/response handling paths (the message conversion/token limit/response extraction/tool parsing blocks that include convert_messages_openai, cJSON building for "messages"/"system", and token keys like "max_tokens"/"max_completion_tokens") so OpenRouter and NVIDIA NIM use the OpenAI-shaped payloads and parsing; update the checks in the blocks currently guarded by provider_is_openai() (around the message-building section and the other sections noted) to use provider_is_openai_compat() instead.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@docs/ARCHITECTURE.md`:
- Around line 57-63: Update the system overview diagram in ARCHITECTURE.md to
include NVIDIA NIM as an LLM provider: add "NVIDIA NIM" to the top provider list
line alongside "Anthropic/OpenAI/OpenRouter API (HTTPS)" and "Brave Search API
(HTTPS)" and ensure the ASCII diagram box or label for "LLM Provider (API
Gateway)" reflects NVIDIA NIM as one of the providers so the visual and text
lists remain consistent.
In `@main/llm/llm_proxy.c`:
- Around line 170-175: llm_api_path() returns the wrong proxy path for
OpenRouter (missing the /api/ prefix) causing proxy requests built in
llm_http_via_proxy() to 404; update llm_api_path() so that when
provider_is_openrouter() is true it returns "/api/v1/chat/completions" (keep
existing "/v1/chat/completions" for OpenAI and NVIDIA Nim and "/v1/messages" as
the default) so proxy-built paths match the full URL returned by llm_api_url().
In `@README.md`:
- Around line 34-35: Update the README text that currently says "NVIDIA NIM
(completely free)" to avoid the inaccurate claim: replace that phrase with a
conditional description such as "NVIDIA NIM (free for development/prototyping
for NVIDIA Developer Program members; production requires NVIDIA AI Enterprise
licensing)" and add a link to NVIDIA's official terms or developer program page;
locate the exact string "NVIDIA NIM (completely free)" in the README and update
it accordingly.
---
Outside diff comments:
In `@main/llm/llm_proxy.c`:
- Around line 552-586: Add a new predicate provider_is_openai_compat() that
returns true when provider_is_openai() or provider_is_openrouter() or
provider_is_nvidia_nim() is true, then replace calls to provider_is_openai()
with provider_is_openai_compat() in all request/response handling paths (the
message conversion/token limit/response extraction/tool parsing blocks that
include convert_messages_openai, cJSON building for "messages"/"system", and
token keys like "max_tokens"/"max_completion_tokens") so OpenRouter and NVIDIA
NIM use the OpenAI-shaped payloads and parsing; update the checks in the blocks
currently guarded by provider_is_openai() (around the message-building section
and the other sections noted) to use provider_is_openai_compat() instead.
In `@main/mimi_secrets.h.example`:
- Around line 20-24: The LLM API section starts a block comment with "/*" that
is never closed, which will comment out subsequent lines including the
MIMI_SECRET_API_KEY, MIMI_SECRET_MODEL, and MIMI_SECRET_MODEL_PROVIDER
definitions; close the block comment by adding the matching "*/" after the
intended comment text (or remove the opening "/*" if the lines are meant to be
active) so the constants (MIMI_SECRET_API_KEY, MIMI_SECRET_MODEL,
MIMI_SECRET_MODEL_PROVIDER) are not accidentally commented out.
---
Duplicate comments:
In `@docs/ARCHITECTURE.md`:
- Around line 282-289: The documentation makes a potentially incorrect claim
that "NVIDIA NIM" is "completely free" in the LLM Provider Integration section;
update the NVIDIA NIM bullet (the string "NVIDIA NIM" / enum key `nvidia_nim`)
to remove or qualify the "completely free" phrase and replace it with accurate,
verifiable wording (e.g., state any free tier limits, trial details, or link to
NVIDIA NIM pricing), ensuring the provider list remains accurate and
non-duplicative.
Member
|
Thanks for your contribution! I will take a look about it. |
Author
|
I changed nvidia_nim to nvidia to avoid issues |
There was a problem hiding this comment.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Duplicate comments:
In `@README.md`:
- Line 43: Replace the misleading "NVIDIA NIM (free)" phrasing in README.md (the
string "NVIDIA NIM (free)") with wording that mirrors NVIDIA's official terms:
state that API access is available free for development/prototyping via the
NVIDIA Developer Program but production use requires NVIDIA AI Enterprise
licensing; update all occurrences (including the other spots noted around lines
181-184) to the same clarified text so the README consistently reflects the
Developer Program vs production licensing distinction.
Author
|
@coderabbitai review |
✅ Actions performedReview triggered.
|
IRONICBo
requested changes
Feb 27, 2026
Member
IRONICBo
left a comment
IRONICBo
left a comment
There was a problem hiding this comment.
Well-structured addition of two OpenAI-compatible providers. The provider_is_openai_compat() abstraction is clean. Issues:
- Merge conflicts — needs rebase onto current main.
- README and ARCHITECTURE.md contain verbose NVIDIA licensing disclaimers repeated 4+ times. Mention it once in the setup section and link to NVIDIA's terms page instead.
- Minor:
llm_api_host()returns hardcoded strings per provider. If custom URL support lands (PR #40), these will diverge. Acceptable for now but worth a TODO comment.
feat: add TODO in llm_proxy.c for deriving host from configured API URL
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
docs/ARCHITECTURE.md (1)
79-85:⚠️ Potential issue | 🟡 MinorUpdate provider-specific stop/tool-call semantics in Data Flow.
Line 79 generalizes to “LLM provider API”, but Lines 81–85 still describe only Anthropic (
stop_reason). OpenAI-compatible providers usefinish_reason(e.g.,tool_calls).📝 Suggested wording update
- iii. If stop_reason == "tool_use": + iii. If Anthropic stop_reason == "tool_use" or OpenAI-compatible finish_reason == "tool_calls": ... - iv. If stop_reason == "end_turn": break with final text + iv. If completion indicates done (Anthropic: "end_turn"; OpenAI-compatible: non-tool-call finish_reason): break with final text🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@docs/ARCHITECTURE.md` around lines 79 - 85, The Data Flow wording currently assumes Anthropic-style stop semantics (stop_reason with values like "tool_use" and "end_turn"); update the description to explicitly cover provider-specific fields by mentioning both Anthropic's stop_reason and OpenAI-style finish_reason (e.g., "tool_calls"/"stop" semantics), and suggest treating them equivalently (map finish_reason=="tool_calls" to stop_reason=="tool_use") so the flow reads: call LLM provider API → parse JSON response for provider-specific stop/finish field (stop_reason or finish_reason) and tool blocks → if provider indicates a tool call (stop_reason == "tool_use" or finish_reason == "tool_calls") execute tools, append results to messages and continue loop → if provider indicates final turn (stop_reason == "end_turn" or finish_reason == "stop") break with final text.
🧹 Nitpick comments (1)
main/llm/llm_proxy.c (1)
248-249: Useprovider_is_openai_compat()consistently in HTTP header branches.Lines 248 and 276 duplicate the compatibility condition manually. Reuse the helper to keep provider logic centralized.
♻️ Suggested refactor
- if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) { + if (provider_is_openai_compat()) { ... - if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) { + if (provider_is_openai_compat()) {Also applies to: 276-277
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@main/llm/llm_proxy.c` around lines 248 - 249, The code duplicates the provider compatibility check by manually combining provider_is_openai(), provider_is_openrouter(), and provider_is_nvidia() where HTTP headers are set; replace those compound conditions with a call to the central helper provider_is_openai_compat() (and remove the old combined checks) so both header branches (the block checking s_api_key and the later block around lines 276-277) consistently use provider_is_openai_compat() to decide OpenAI-compatible header behavior.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@docs/ARCHITECTURE.md`:
- Around line 236-237: The docs claim there is "no runtime configuration" but
the code actually supports runtime overrides via CLI and NVS; update the
ARCHITECTURE.md text around the configuration section so it no longer asserts
"no runtime configuration" and instead documents that environment variables like
MIMI_SECRET_API_KEY and MIMI_SECRET_MODEL can be overridden at runtime using the
CLI and NVS (include brief notes on precedence and default model value
claude-opus-4-6); ensure the table entries for MIMI_SECRET_API_KEY and
MIMI_SECRET_MODEL remain but add a sentence referencing the supported runtime
overrides (CLI/NVS) and the override precedence.
In `@main/llm/llm_proxy.c`:
- Around line 147-150: The provider_is_nvidia function only checks for "nvidia"
and breaks compatibility with legacy "nvidia_nim"; update provider_is_nvidia to
treat both "nvidia" and "nvidia_nim" as NVIDIA providers (e.g., check
strcmp(s_provider, "nvidia")==0 || strcmp(s_provider, "nvidia_nim")==0 or test a
prefix like strncmp(s_provider, "nvidia", 6)==0) so routing uses the NVIDIA
URL/headers when s_provider holds either value; modify the provider_is_nvidia
function accordingly.
---
Outside diff comments:
In `@docs/ARCHITECTURE.md`:
- Around line 79-85: The Data Flow wording currently assumes Anthropic-style
stop semantics (stop_reason with values like "tool_use" and "end_turn"); update
the description to explicitly cover provider-specific fields by mentioning both
Anthropic's stop_reason and OpenAI-style finish_reason (e.g.,
"tool_calls"/"stop" semantics), and suggest treating them equivalently (map
finish_reason=="tool_calls" to stop_reason=="tool_use") so the flow reads: call
LLM provider API → parse JSON response for provider-specific stop/finish field
(stop_reason or finish_reason) and tool blocks → if provider indicates a tool
call (stop_reason == "tool_use" or finish_reason == "tool_calls") execute tools,
append results to messages and continue loop → if provider indicates final turn
(stop_reason == "end_turn" or finish_reason == "stop") break with final text.
---
Nitpick comments:
In `@main/llm/llm_proxy.c`:
- Around line 248-249: The code duplicates the provider compatibility check by
manually combining provider_is_openai(), provider_is_openrouter(), and
provider_is_nvidia() where HTTP headers are set; replace those compound
conditions with a call to the central helper provider_is_openai_compat() (and
remove the old combined checks) so both header branches (the block checking
s_api_key and the later block around lines 276-277) consistently use
provider_is_openai_compat() to decide OpenAI-compatible header behavior.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: f0fdbe8a-2c2f-4b17-a507-a36c67a6ead8
📒 Files selected for processing (4)
README.mddocs/ARCHITECTURE.mdmain/cli/serial_cli.cmain/llm/llm_proxy.c
🚧 Files skipped from review as they are similar to previous changes (1)
- main/cli/serial_cli.c
Comment on lines
+236
to
237
| | `MIMI_SECRET_API_KEY` | LLM API key (Anthropic, OpenAI, OpenRouter, or NVIDIA NIM) | | ||
| | `MIMI_SECRET_MODEL` | Model ID (default: claude-opus-4-6) | |
There was a problem hiding this comment.
Configuration section still conflicts with runtime override behavior.
Around this section, Line 229 states there is no runtime configuration, but current behavior uses CLI + NVS overrides. This will mislead operators.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@docs/ARCHITECTURE.md` around lines 236 - 237, The docs claim there is "no
runtime configuration" but the code actually supports runtime overrides via CLI
and NVS; update the ARCHITECTURE.md text around the configuration section so it
no longer asserts "no runtime configuration" and instead documents that
environment variables like MIMI_SECRET_API_KEY and MIMI_SECRET_MODEL can be
overridden at runtime using the CLI and NVS (include brief notes on precedence
and default model value claude-opus-4-6); ensure the table entries for
MIMI_SECRET_API_KEY and MIMI_SECRET_MODEL remain but add a sentence referencing
the supported runtime overrides (CLI/NVS) and the override precedence.
Comment on lines
+147
to
+150
| static bool provider_is_nvidia(void) | ||
| { | ||
| return strcmp(s_provider, "nvidia") == 0; | ||
| } |
There was a problem hiding this comment.
Preserve backward compatibility for prior nvidia_nim provider values.
Line 149 only matches "nvidia". If NVS/build-time config still contains "nvidia_nim", routing falls back to Anthropic URL/headers and requests fail.
🔧 Suggested fix
static bool provider_is_nvidia(void)
{
- return strcmp(s_provider, "nvidia") == 0;
+ return strcmp(s_provider, "nvidia") == 0 ||
+ strcmp(s_provider, "nvidia_nim") == 0; /* backward-compatible alias */
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| static bool provider_is_nvidia(void) | |
| { | |
| return strcmp(s_provider, "nvidia") == 0; | |
| } | |
| static bool provider_is_nvidia(void) | |
| { | |
| return strcmp(s_provider, "nvidia") == 0 || | |
| strcmp(s_provider, "nvidia_nim") == 0; /* backward-compatible alias */ | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@main/llm/llm_proxy.c` around lines 147 - 150, The provider_is_nvidia function
only checks for "nvidia" and breaks compatibility with legacy "nvidia_nim";
update provider_is_nvidia to treat both "nvidia" and "nvidia_nim" as NVIDIA
providers (e.g., check strcmp(s_provider, "nvidia")==0 || strcmp(s_provider,
"nvidia_nim")==0 or test a prefix like strncmp(s_provider, "nvidia", 6)==0) so
routing uses the NVIDIA URL/headers when s_provider holds either value; modify
the provider_is_nvidia function accordingly.
…portal, SPIFFS fix - Resolved conflict in main/llm/llm_proxy.c: removed dead llm_chat() backward-compat stub (not declared in header, not referenced); NVIDIA NIM support retained - Resolved conflict in main/mimi_secrets.h.example: kept upstream Feishu Bot section plus fork's descriptive LLM provider comment
There was a problem hiding this comment.
♻️ Duplicate comments (1)
main/llm/llm_proxy.c (1)
195-198:⚠️ Potential issue | 🟠 MajorPreserve backward compatibility for
nvidia_nimprovider value.The function only matches
"nvidia", but existing NVS configurations may contain"nvidia_nim"from before the rename. These configurations would silently fall back to Anthropic routing, causing authentication failures.🔧 Proposed fix
static bool provider_is_nvidia(void) { - return strcmp(s_provider, "nvidia") == 0; + return strcmp(s_provider, "nvidia") == 0 || + strcmp(s_provider, "nvidia_nim") == 0; }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@main/llm/llm_proxy.c` around lines 195 - 198, provider_is_nvidia currently only returns true for s_provider == "nvidia", which breaks backward compatibility for the legacy "nvidia_nim" value; update provider_is_nvidia to treat both "nvidia" and "nvidia_nim" as NVIDIA providers (e.g., check strcmp(s_provider, "nvidia") == 0 || strcmp(s_provider, "nvidia_nim") == 0) so existing NVS configs continue to route correctly while preserving the s_provider variable and function name provider_is_nvidia.
🧹 Nitpick comments (2)
main/llm/llm_proxy.c (2)
324-343: Same refactor opportunity applies here.This inline provider check mirrors the one in
llm_http_direct(). Consider usingprovider_is_openai_compat()for both locations.♻️ Proposed refactor
- if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) { + if (provider_is_openai_compat()) { hlen = snprintf(header, sizeof(header),🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@main/llm/llm_proxy.c` around lines 324 - 343, The header-construction block in llm_proxy.c duplicates the provider check used elsewhere; replace the compound check (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) with the single helper provider_is_openai_compat() to centralize provider logic, keeping the two snprintf branches and the same arguments (llm_api_path(), llm_api_host(), s_api_key, body_len or MIMI_LLM_API_VERSION) intact; this mirrors the refactor already applied in llm_http_direct() and prevents divergent provider logic.
296-305: Consider usingprovider_is_openai_compat()for consistency.The inline check duplicates the logic in
provider_is_openai_compat()which is used elsewhere in this file. Using the helper would improve maintainability if provider grouping changes.♻️ Proposed refactor
- if (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) { + if (provider_is_openai_compat()) { if (s_api_key[0]) {🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@main/llm/llm_proxy.c` around lines 296 - 305, Replace the inline provider check (provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) with the helper provider_is_openai_compat() to centralize provider grouping; keep the existing behavior: when provider_is_openai_compat() is true and s_api_key is non-empty set the "Authorization" header to "Bearer <s_api_key>" (using the existing auth buffer logic referencing LLM_API_KEY_MAX_LEN and esp_http_client_set_header), otherwise set "x-api-key" to s_api_key and "anthropic-version" to MIMI_LLM_API_VERSION so the rest of the code (esp_http_client_set_header, s_api_key, MIMI_LLM_API_VERSION) continues to work as before.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Duplicate comments:
In `@main/llm/llm_proxy.c`:
- Around line 195-198: provider_is_nvidia currently only returns true for
s_provider == "nvidia", which breaks backward compatibility for the legacy
"nvidia_nim" value; update provider_is_nvidia to treat both "nvidia" and
"nvidia_nim" as NVIDIA providers (e.g., check strcmp(s_provider, "nvidia") == 0
|| strcmp(s_provider, "nvidia_nim") == 0) so existing NVS configs continue to
route correctly while preserving the s_provider variable and function name
provider_is_nvidia.
---
Nitpick comments:
In `@main/llm/llm_proxy.c`:
- Around line 324-343: The header-construction block in llm_proxy.c duplicates
the provider check used elsewhere; replace the compound check
(provider_is_openai() || provider_is_openrouter() || provider_is_nvidia()) with
the single helper provider_is_openai_compat() to centralize provider logic,
keeping the two snprintf branches and the same arguments (llm_api_path(),
llm_api_host(), s_api_key, body_len or MIMI_LLM_API_VERSION) intact; this
mirrors the refactor already applied in llm_http_direct() and prevents divergent
provider logic.
- Around line 296-305: Replace the inline provider check (provider_is_openai()
|| provider_is_openrouter() || provider_is_nvidia()) with the helper
provider_is_openai_compat() to centralize provider grouping; keep the existing
behavior: when provider_is_openai_compat() is true and s_api_key is non-empty
set the "Authorization" header to "Bearer <s_api_key>" (using the existing auth
buffer logic referencing LLM_API_KEY_MAX_LEN and esp_http_client_set_header),
otherwise set "x-api-key" to s_api_key and "anthropic-version" to
MIMI_LLM_API_VERSION so the rest of the code (esp_http_client_set_header,
s_api_key, MIMI_LLM_API_VERSION) continues to work as before.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 64567f3d-2047-4f1a-9d47-f6f6a9e70cf6
📒 Files selected for processing (6)
README.mdmain/cli/serial_cli.cmain/llm/llm_proxy.cmain/llm/llm_proxy.hmain/mimi_config.hmain/mimi_secrets.h.example
✅ Files skipped from review due to trivial changes (1)
- main/llm/llm_proxy.h
🚧 Files skipped from review as they are similar to previous changes (2)
- main/cli/serial_cli.c
- main/mimi_secrets.h.example
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add OpenRouter & NVIDIA NIM LLM Providers
Summary
Add OpenRouter and NVIDIA NIM (completely free) as LLM providers alongside Anthropic and OpenAI.
Both expose OpenAI-compatible chat-completions endpoints, reusing existing OpenAI request/response logic with no new serialization code needed.
Architecture
llm_api_url()llm_api_host()llm_api_path()Usage
OpenRouter
NVIDIA NIM (free)
Available Models
OpenRouter
anthropic/claude-sonnet-4.6anthropic/claude-opus-4.6openai/gpt-5.2-proopenai/gpt-5.2openai/gpt-5.1google/gemini-3-pro-previewgoogle/gemini-3-flash-previewqwen/qwen3.5-397b-a17bdeepseek/deepseek-v3.2x-ai/grok-4.1-fastFree models:
openrouter/freeopenrouter/gpt-oss-120b:freeNVIDIA NIM (Completely Free)
qwen/qwen3.5-397b-a17bz-ai/glm5nvidia/embeddingminimaxai/minimax-m2.1stepfun-ai/step-3.5-flashmoonshotai/kimi-2.5Get an API key: https://org.ngc.nvidia.com/setup/api-keys
What Changed
main/mimi_config.hMIMI_OPENROUTER_API_URLandMIMI_NVIDIA_NIM_API_URLmain/llm/llm_proxy.cprovider_is_openrouter()/provider_is_nvidia_nim()and update routingmain/llm/llm_proxy.hmain/cli/serial_cli.cmain/mimi_secrets.h.exampleREADME.mddocs/ARCHITECTURE.mdTest Plan
idf.py buildset_model_provider openrouteris accepted via serial CLIset_model_provider nvidia_nimis accepted via serial CLIset_api_keycorrectly stores the OpenRouter/NVIDIA NIM API key in NVSanthropicoropenaicontinues to work without regressionSummary by CodeRabbit
New Features
Documentation