DOC-1867 AI Gateway

[micheleRP]

Description

This pull request introduces the AI Gateway in Redpanda Cloud.

Resolves https://redpandadata.atlassian.net/browse/DOC-1867
Review deadline: Feb 6

Page previews

• Overview
• Quickstart
• Architecture
• Setup Guide
• Discover Gateways
• Connect Your Agent
• CEL Routing Cookbook
• MCP Aggregation
• Integrations

Checks

• New feature
• Content gap
• Support Follow-up
• Small fix (typos, links, copyedits, etc)

[coderabbitai]

Important

Review skipped

Auto incremental reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

• 🔍 Trigger a full review

📝 Walkthrough

Walkthrough

Adds extensive AI Gateway documentation and navigation: new AI Gateway index, overview, quickstart, architecture, CEL routing cookbook, observability (logs/metrics), migration guide, MCP aggregation/orchestration guide, administrator and builder setup guides, discovery/connect guides, many IDE/tool integration pages (Claude Code, Cline, Continue.dev, Cursor, GitHub Copilot), MCP Server Patterns/nav updates, a BYOC support partial, and renames the AI Agents index to "Agentic AI". All changes are documentation-only.

Sequence Diagram(s)

sequenceDiagram
participant Client
participant AI_Gateway as "AI Gateway"
participant MCP_Server as "MCP Server"
participant LLM_Provider as "LLM Provider"
participant Observability as "Observability"

Client->>AI_Gateway: Send request (headers, rp-aigw-id, body)
AI_Gateway->>Observability: Log request + metrics
AI_Gateway->>AI_Gateway: Apply policies (CEL, rate/spend limits, routing)
AI_Gateway->>MCP_Server: Discover/Invoke tools (deferred load / orchestrator)
MCP_Server-->>AI_Gateway: Tool result(s)
AI_Gateway->>LLM_Provider: Forward transformed request (provider pool / fallback)
LLM_Provider-->>AI_Gateway: Response (tokens, cost)
AI_Gateway->>Observability: Log response, costs, routing decisions
AI_Gateway-->>Client: Consolidated response (LLM + MCP tool outputs)

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• Add docs for remote MCP servers #413: Overlaps on AI Agents/MCP navigation edits and MCP-remote entries.
• Improve MCP devex #478: Related MCP navigation and MCP Server Patterns changes.
• Reorganize navigation structure and rename AI Agents to Agentic AI #490: Similar AI Agents and MCP navigation restructuring affecting same nav.adoc area.

Suggested reviewers

• birdayz
• paulohtb6
• JakeSCahill

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The PR title 'DOC-1867 AI Gateway' directly references the linked issue and names the feature being documented, providing sufficient context about the main change.
Description check	✅ Passed	The PR description follows the template structure, includes the linked issue reference (DOC-1867), a clear review deadline (Feb 6), comprehensive page previews for all documentation pages, and correctly marks the PR as a new feature.
Linked Issues check	✅ Passed	The PR introduces comprehensive AI Gateway documentation covering overview, quickstart, architecture, setup guide, integrations, routing, and MCP aggregation, fully addressing the requirement to document the AI Gateway feature and support the cloud team.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to documenting the AI Gateway feature. The navigation update, partial addition, and index page modification are necessary supporting changes for the documentation structure.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[netlify]

✅ Deploy Preview for rp-cloud ready!

Name	Link
🔨 Latest commit	000b819
🔍 Latest deploy log	https://app.netlify.com/projects/rp-cloud/deploys/697b9aa054b041000834f872
😎 Deploy Preview	https://deploy-preview-477--rp-cloud.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[c4milo]

it is an MCP server exposing tools, so it is used as another regular MCP server. I can provide examples.

[c4milo]

I haven't gotten to the observability implementation yet, probably next week.

[c4milo]

It is not implemented yet.

[c4milo]

This configuration should use http as transport. For example, to configure our internal MCP gateway in Claude Code, it would look roughly like this:

claude mcp add --transport http redpanda-aigateway https://gw.ai.panda.dev/mcp --header "Authorization: Bearer YOUR_API_KEY" --header "rp-aigw-id: GATEWAY_ID"

[c4milo]

There is also a way to disallow people from using other MCP servers. I'm going to have to dedicate some time to documenting and testing all the integrations with popular AI coding agents.

[micheleRP]

@coderabbitai review

[coderabbitai]

✅ Actions performed

Review triggered.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 9

🤖 Fix all issues with AI agents

In `@modules/ai-agents/pages/ai-gateway/integrations/claude-code-admin.adoc`:
- Around line 313-355: Update the example config to match Anthropic's MCP
schema: change the file path from "~/.claude/config.json" to the correct
user/state file "~/.claude.json" (or mention project-level ".mcp.json"), remove
the undocumented "apiProviders" section entirely, and update the "mcpServers"
entry to use "type": "http" instead of "transport" (i.e., replace the
"transport" key with "type" and keep "url" and "headers" under the
"redpanda-ai-gateway" object).

In `@modules/ai-agents/pages/ai-gateway/integrations/claude-code-user.adoc`:
- Around line 184-210: The apiProviders example incorrectly shows environment
variable interpolation which only works in the mcpServers section; update the
document so apiProviders does not present "${REDPANDA_GATEWAY_URL}" or
"${REDPANDA_GATEWAY_ID}" as valid interpolation options—either replace those
with sample hardcoded values or add a clear note that only mcpServers supports
${VAR} interpolation and apiProviders requires explicit values or a different
dynamic config mechanism; reference the mcpServers and apiProviders sections and
the REDPANDA_GATEWAY_URL / REDPANDA_GATEWAY_ID symbols so readers know which
fields support interpolation.

In `@modules/ai-agents/pages/ai-gateway/integrations/cline-admin.adoc`:
- Around line 463-471: Replace the token‑like literal in the curl example header
to use an environment variable placeholder instead of "YOUR_API_TOKEN"; update
the Authorization header in the shown curl (the line with -H "Authorization:
Bearer YOUR_API_TOKEN") to something like -H "Authorization: Bearer
$REDPANDA_API_TOKEN" (or use shell expansion via -H "Authorization: Bearer
${REDPANDA_API_TOKEN}") and add a short comment or note that the env var must be
set, so secret scanners won't flag the example.
- Around line 340-383: Remove the unofficial VS Code settings keys
("cline.apiProvider", "cline.apiBaseUrl", "cline.customHeaders") from the
settings.json example and instead document that API/provider configuration is
managed via the Cline extension global state (not VS Code settings); replace the
MCP server example that used "settings.json" and "transport" with an example
that references cline_mcp_settings.json and uses the official fields (use "type"
e.g., "streamableHttp" instead of "transport" and include the proper "url" and
"headers" structure per Cline's remote server schema); also mention the VS Code
toggle name "Cline > Mcp: Mode" as the global MCP enablement setting and
instruct users to configure MCP servers via the Cline UI or
cline_mcp_settings.json rather than settings.json.

In `@modules/ai-agents/pages/ai-gateway/integrations/continue-admin.adoc`:
- Around line 422-517: Update the examples to the current Continue.dev YAML
config standard by renaming the file and examples from ~/.continue/config.json
to ~/.continue/config.yaml and converting the JSON examples to equivalent YAML;
replace the MCP block key experimental.modelContextProtocolServers (plural) with
experimental.modelContextProtocolServer (singular) and ensure each
transport.type uses a valid value (stdio, sse, or streamable-http) instead of
"http"; update apiBase and model/provider entries to YAML syntax (references:
"models", "tabAutocompleteModel", "apiBase", "apiKey") and add guidance to
prefer the .continue/mcpServers/ directory-based MCP configuration method rather
than embedding MCP servers in the main config.yaml.

In `@modules/ai-agents/pages/ai-gateway/integrations/continue-user.adoc`:
- Around line 269-324: The config examples use legacy config.json interpolation
like "${REDPANDA_API_KEY}" which Continue.dev does not expand; replace the JSON
example with an equivalent config.yaml example that uses the correct
interpolation syntax `${{ secrets.REDPANDA_API_KEY }}`, `${{
secrets.REDPANDA_GATEWAY_URL }}`, and `${{ secrets.REDPANDA_GATEWAY_ID }}` for
the keys referenced in the snippet (models -> apiKey/apiBase,
requestOptions.headers.rp-aigw-id, experimental.modelContextProtocolServers ->
transport.url and transport.headers.Authorization/rp-aigw-id), or explicitly
mark config.json as deprecated and add a short note that `${VAR}` will be
treated as a literal string and provide the config.yaml alternative (using the
same fields: models, provider/model, apiKey, apiBase, requestOptions,
experimental.modelContextProtocolServers, transport) so users know to use `${{
secrets.* }}` in config.yaml.

In `@modules/ai-agents/pages/ai-gateway/integrations/cursor-user.adoc`:
- Around line 193-221: The settings.json snippet uses unsupported ${VAR}
interpolation (for keys cursor.overrideOpenAIBaseUrl,
cursor.overrideOpenAIApiKey, and openai.additionalHeaders) so those placeholders
will remain literal; update the doc to either replace the ${VAR} placeholders
with concrete values that users must paste into settings.json, or use the
supported environment-syntax if available (e.g., ${env:VAR}) for those specific
settings, or recommend generating the settings.json at runtime via a small setup
script that reads REDPANDA_GATEWAY_URL, REDPANDA_GATEWAY_ID, and
REDPANDA_API_KEY and writes the actual values into cursor.overrideOpenAIBaseUrl,
cursor.overrideOpenAIApiKey, and openai.additionalHeaders before launching
Cursor.

In `@modules/ai-agents/pages/ai-gateway/integrations/github-copilot-user.adoc`:
- Around line 86-217: The documented settings are incorrect and must be
replaced: change references of github.copilot.advanced.customModels to
github.copilot.chat.customOAIModels (experimental chat OAI models) and remove
the non-existent useOAIProvider usage; replace oai.provider.* examples with the
OAI Compatible Provider extension's actual settings and flow—use
oaicopilot.baseUrl and oaicopilot.models where appropriate and note that API
keys/headers are set via the Copilot Chat → Manage Models UI (not in
settings.json); update the examples and prose to instruct users to open Copilot
Chat → Manage Models to add the OAI Compatible provider, enter base URL, api key
and custom headers there, and then select the configured model in the Copilot
UI.

In `@modules/ai-agents/pages/ai-gateway/migration-guide.adoc`:
- Around line 623-638: The example config duplicates the rp-aigw-id by placing
it in additional_kwargs (which lands in the request body) and default_headers
(which sets HTTP headers); update the OpenAI client construction inside the
use_gateway branch so rp-aigw-id is only set via default_headers and remove it
from additional_kwargs—locate the OpenAI(...) call (symbol: OpenAI) assigned to
llm in the use_gateway block and delete the rp-aigw-id entry from
additional_kwargs while keeping default_headers={"rp-aigw-id":
os.getenv("REDPANDA_AI_GATEWAY_ID")}.

♻️ Duplicate comments (4)

modules/ai-agents/pages/ai-gateway/observability-logs.adoc (1)

31-731: Resolve remaining PLACEHOLDER sections before publish.
This doc still contains multiple “PLACEHOLDER” markers (UI paths, screenshots, retention/export, integrations, redaction, RBAC, etc.). These are core reference points and should be finalized or explicitly scoped as internal/TBD.

modules/ai-agents/pages/ai-gateway/observability-metrics.adoc (1)

31-849: Resolve remaining PLACEHOLDER sections before publish.
This page still has multiple “PLACEHOLDER” markers (UI paths, screenshots, alerting/export details, integrations, retention, delays). These should be finalized or explicitly marked as internal/TBD.

modules/ai-agents/pages/ai-gateway/migration-guide.adoc (1)

362-370: Note: Product-specific placeholders remain unresolved.

Multiple placeholder sections still need to be filled with concrete values:

• UI navigation paths (line 362, 741)
• Expected gateway overhead/latency (lines 440-444, 770)
• Model catalog links (line 743)
• Log ingestion delay (line 787)
• Response format differences (lines 794-795)
• Streaming support confirmation (line 829)

These were previously flagged in past review comments. Consider either finalizing these values with product/engineering confirmation or marking them as "TBD" with a docs issue tracker.

Also applies to: 440-444, 741-743, 770-770, 787-787, 794-795, 829-829

modules/ai-agents/pages/ai-gateway/mcp-aggregation-guide.adoc (1)

288-310: Note: MCP configuration and operational placeholders remain unresolved.

Multiple placeholder sections still need finalization:

• MCP configuration format (UI paths, gateway-level vs per-request deferred loading)
• Orchestrator API security limits (timeout, memory, other constraints)
• MCP server approval workflow details
• MCP server versioning strategy
• MCP-specific metrics availability and dashboard details
• Sandboxing implementation specifics
• LangChain-specific integration code

These were flagged in previous review. Consider finalizing with product/engineering input or marking as "Internal draft - details TBD" until confirmed.

Also applies to: 443-474, 560-621, 664-691, 744-761, 812-821, 1008-1008

🧹 Nitpick comments (1)

modules/ai-agents/pages/ai-gateway/mcp-aggregation-guide.adoc (1)

936-970: Consider enhancing the tool call handling example.

The Python integration example shows a basic tool call pattern, but it could be improved for production use:

1. Incomplete message history (lines 959-970): The follow-up completion after tool execution only includes three messages (user, assistant, tool result). For multi-turn tool usage, you should include the full conversation history or at least maintain context across tool calls.

2. Missing tools parameter in follow-up call (line 959): The second client.chat.completions.create() call doesn't include the tools parameter, preventing the model from making additional tool calls if needed for multi-step workflows.

💡 Enhanced example with proper message history

 # Agent uses tools
+messages = [{"role": "user", "content": "Query the database for premium users"}]
+
 response = client.chat.completions.create(
     model="anthropic/claude-sonnet-3.5",
-    messages=[
-        {"role": "user", "content": "Query the database for premium users"}
-    ],
+    messages=messages,
     tools=tools,
     tool_choice="auto"
 )

 # Handle tool calls
 if response.choices[0].message.tool_calls:
+    messages.append(response.choices[0].message)
+    
     for tool_call in response.choices[0].message.tool_calls:
         # Execute tool via gateway
         tool_result = requests.post(
             f"https://{CLUSTER_ID}.cloud.redpanda.com/ai-gateway/mcp/tools/{tool_call.function.name}",
             headers={
                 "Authorization": f"Bearer {os.getenv('REDPANDA_CLOUD_TOKEN')}",
                 "rp-aigw-id": os.getenv("GATEWAY_ID")
             },
             json=json.loads(tool_call.function.arguments)
         )

-        # Continue conversation with tool result
+        messages.append({
+            "role": "tool",
+            "tool_call_id": tool_call.id,
+            "content": json.dumps(tool_result.json())
+        })
+
+    # Continue conversation with full history and tools available
-        response = client.chat.completions.create(
-            model="anthropic/claude-sonnet-3.5",
-            messages=[
-                {"role": "user", "content": "Query the database for premium users"},
-                response.choices[0].message,
-                {
-                    "role": "tool",
-                    "tool_call_id": tool_call.id,
-                    "content": json.dumps(tool_result.json())
-                }
-            ]
-        )
+    response = client.chat.completions.create(
+        model="anthropic/claude-sonnet-3.5",
+        messages=messages,
+        tools=tools,  # Include tools for potential follow-up calls
+        tool_choice="auto"
+    )

[micheleRP]

@coderabbitai review

[coderabbitai]

✅ Actions performed

Review triggered.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

✅ Actions performed

Review triggered.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 3

🤖 Fix all issues with AI agents

In `@modules/ai-agents/pages/ai-gateway/integrations/continue-admin.adoc`:
- Around line 659-663: The doc incorrectly references `config.json` while
Continue.dev uses `config.yaml`; update all occurrences in this section so
examples and warnings refer to `config.yaml` (e.g., change "`config.json`" to
"`config.yaml`" and update the chmod example to `~/.continue/config.yaml`), and
keep the guidance points the same: never commit the config file, restrict
filesystem permissions, and rotate tokens if compromised.

In `@modules/ai-agents/pages/ai-gateway/integrations/continue-user.adoc`:
- Around line 309-324: The example .continuerc.json uses unsupported ${VAR}
interpolation (e.g., "${PROJECT_API_KEY}" and "${PROJECT_GATEWAY_ID}") which
will be treated as literal strings; update the `.continuerc.json` example to use
explicit placeholder text (like PROJECT_API_KEY_PLACEHOLDER) or switch the docs
to show using a supported format (recommend using config.yaml with ${{ secrets.*
}} interpolation or config.ts for programmatic env access) and mention that
config.json does not support ${VAR} expansion to avoid confusion.
- Around line 285-292: Replace the incorrect YAML key
experimental.modelContextProtocolServer with the correct config.yaml schema key
mcpServers and convert the single object into an array entry; update the block
so mcpServers is an array of server objects each containing transport, url and
headers (keeping transport.type, transport.url and headers.Authorization and
headers.rp-aigw-id) to match the structure used for modelContextProtocolServers
in the JSON example; ensure the unique symbols updated are
experimental.modelContextProtocolServer -> mcpServers and transport/url/headers
remain under the array element.

♻️ Duplicate comments (1)

modules/ai-agents/pages/ai-gateway/integrations/github-copilot-user.adoc (1)

274-278: Correct invalid configuration settings—oai.provider.* settings do not exist in the OAI Compatible Provider extension.

Multiple sections use oai.provider.* settings (e.g., oai.provider.headers, oai.provider.endpoint, oai.provider.apiKey, oai.provider.models, oai.provider.timeout, oai.provider.debug) which do not exist in the OAI Compatible Provider extension's documented API. Based on the past review and official documentation, the extension uses:

• oaicopilot.baseUrl (not oai.provider.endpoint)
• oaicopilot.models (not oai.provider.models)
• API keys and custom headers are configured via Copilot Chat → Manage Models UI, not in settings.json

These incorrect settings will cause non-functional configurations for users.

🔧 Required corrections

Lines 274-278, 294-299 (Multi-tenancy, Approach 1):

 {
-  "oai.provider.headers": {
+  // Note: Headers must be configured via Copilot Chat → Manage Models UI
+  // This example shows the structure for documentation purposes only
+  "oaicopilot.baseUrl": "https://gw.ai.panda.com/v1",
+  "customHeaders": {
     "rp-aigw-id": "team-a-gateway-123"
   }
 }

Lines 367-375 (Centralized configuration):

 {
-  "oai.provider.endpoint": "https://gw.company.com/v1",
-  "oai.provider.apiKey": "${env:COPILOT_GATEWAY_KEY}",
-  "oai.provider.headers": {
+  "oaicopilot.baseUrl": "https://gw.company.com/v1",
+  // Note: API keys and headers must be configured via Copilot Chat → Manage Models UI
+  // Environment variables shown here for reference only
+  "headers": {
     "rp-aigw-id": "${env:COPILOT_GATEWAY_ID}",
     "x-user-email": "${env:USER_EMAIL}",
     "x-department": "${env:DEPARTMENT}"
   }
 }

Add a note in each section explaining that while oaicopilot.baseUrl and oaicopilot.models can be set in settings.json, API keys and custom headers must be configured through the Copilot Chat UI (Manage Models).

Also applies to: 367-375, 505-522, 674-678, 769-774

🧹 Nitpick comments (6)

modules/ai-agents/pages/ai-gateway/integrations/cursor-user.adoc (1)

546-551: Consider using environment variable syntax in curl examples to avoid triggering security scanners.

The placeholder YOUR_API_KEY in the curl command may trigger secret detection tools. While this is a documentation example, using environment variable syntax (e.g., ${REDPANDA_API_KEY} or $REDPANDA_API_KEY) would better demonstrate secure credential handling and avoid false positives from security scanners.

♻️ Suggested improvement

 [,bash]
 ----
-curl -H "Authorization: Bearer YOUR_API_KEY" \
+curl -H "Authorization: Bearer ${REDPANDA_API_KEY}" \
      -H "rp-aigw-id: GATEWAY_ID" \
      https://gw.ai.panda.com/v1/models
 ----

modules/ai-agents/pages/ai-gateway/integrations/claude-code-user.adoc (1)

307-312: Consider using environment variable syntax in curl examples for consistency.

Similar to other integration docs, consider using environment variable syntax (${REDPANDA_API_KEY}) instead of placeholder text to demonstrate secure credential handling and maintain consistency across the documentation.

♻️ Suggested improvement

 [,bash]
 ----
-curl -H "Authorization: Bearer YOUR_API_KEY" \
+curl -H "Authorization: Bearer ${REDPANDA_API_KEY}" \
      -H "rp-aigw-id: GATEWAY_ID" \
      https://gw.ai.panda.com/mcp
 ----

modules/ai-agents/pages/ai-gateway/integrations/github-copilot-user.adoc (1)

632-637: Use environment variable syntax in curl authentication examples.

The placeholder YOUR_API_KEY may trigger security scanners. Use environment variable syntax to demonstrate secure credential handling.

♻️ Suggested improvement

 [,bash]
 ----
-curl -H "Authorization: Bearer YOUR_API_KEY" \
+curl -H "Authorization: Bearer ${REDPANDA_API_KEY}" \
      -H "rp-aigw-id: GATEWAY_ID" \
      https://gw.ai.panda.com/v1/models
 ----

Also applies to: 672-678

modules/ai-agents/pages/ai-gateway/integrations/claude-code-admin.adoc (1)

402-413: Consider using environment variable syntax in curl examples.

For consistency with security best practices across the documentation, consider using environment variable syntax in the curl example.

♻️ Suggested improvement

 [source,bash]
 ----
+# Set REDPANDA_API_TOKEN environment variable before running
 curl https://{CLUSTER_ID}.cloud.redpanda.com/api/ai-gateway/logs \
-  -H "Authorization: Bearer YOUR_API_TOKEN" \
+  -H "Authorization: Bearer ${REDPANDA_API_TOKEN}" \
   -H "Content-Type: application/json" \
   -d '{
     "gateway_id": "GATEWAY_ID",
     "start_time": "2026-01-01T00:00:00Z",
     "end_time": "2026-01-14T23:59:59Z",
     "limit": 100
   }'
 ----

modules/ai-agents/pages/ai-gateway/integrations/continue-admin.adoc (1)

608-617: Use environment variable syntax in curl example for security best practices.

For consistency with other admin documentation and to avoid triggering security scanners, use environment variable syntax in the curl authentication header.

♻️ Suggested improvement

 [source,bash]
 ----
+# Set REDPANDA_API_TOKEN environment variable before running
 curl https://{CLUSTER_ID}.cloud.redpanda.com/api/ai-gateway/logs \
-  -H "Authorization: Bearer YOUR_API_TOKEN" \
+  -H "Authorization: Bearer ${REDPANDA_API_TOKEN}" \
   -H "Content-Type: application/json" \
   -d '{
     "gateway_id": "GATEWAY_ID",
     "start_time": "2026-01-01T00:00:00Z",
     "end_time": "2026-01-14T23:59:59Z",
     "limit": 100
   }'
 ----

modules/ai-agents/pages/ai-gateway/integrations/continue-user.adoc (1)

592-596: Prefer env var in curl example to avoid shell-history leaks.

Even with placeholders, examples like this are commonly copied verbatim. Consider using an env var in the snippet.

✅ Suggested tweak

-curl -H "Authorization: Bearer YOUR_API_KEY" \
+curl -H "Authorization: Bearer ${REDPANDA_API_KEY}" \
      -H "rp-aigw-id: GATEWAY_ID" \
      https://gw.ai.panda.com/v1/models