BlueFish AIAQI AI Intelligence
⌕

BLUEFISH / STACK

Stack Lab

Research and guides only. This website does not connect your accounts or remotely control devices.

A missing exact price does not mean a service is free. Regional access not tested

DeepSeek Harness × Ollama

Official support

Use Ollama’s official Harness launcher and choose an installed model.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Review official release notes and prepare Node.js, Ollama and an installed model; record versions.
  2. Read Ollama’s official Harness integration and use --config first to inspect configuration behavior.
  3. Launch dsh with the chosen model, inspect the actual provider/model and start with a read-only task.
  4. Inspect web-search settings; if local-only is required, disable cloud features and inspect network records.
  5. Test tool use, cancellation and recovery in a test directory and retain human approvals.
  • Developer preview; tools depend on the model and web search uses the cloud.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Ollama with Harness ↗DeepSeek Harness official documentation ↗

DeepSeek Harness × LM Studio

Partial compatibility

Investigate Custom model API using a shared LM Studio protocol; only both endpoint documents were reviewed.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Open the official sources below and record installed versions, model ID and current config; no end-to-end tested version is pinned here.
  2. Investigate Custom model API using a shared LM Studio protocol; only both endpoint documents were reviewed.
  3. Use only capabilities documented at both ends; verify endpoint and model before roles, streaming and tool fields, without assuming full compatibility.
  4. Use the source configuration page to select the actual provider and protocol, then send a short read-only request in isolation; do not guess missing parameters.
  5. Verify text, needed tools, cancellation, timeouts and recovery separately; save results and limits, revert on failure and retain the untested label.
  • Test tool calls, system roles, streaming and token fields; this is not an end-to-end support claim.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Harness model configuration ↗LM Studio official documentation ↗

DeepSeek Harness × vLLM

Partial compatibility

Investigate a custom OpenAI Chat Completions provider with vLLM serving.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Open the official sources below and record installed versions, model ID and current config; no end-to-end tested version is pinned here.
  2. Investigate a custom OpenAI Chat Completions provider with vLLM serving.
  3. Use only capabilities documented at both ends; verify endpoint and model before roles, streaming and tool fields, without assuming full compatibility.
  4. Use the source configuration page to select the actual provider and protocol, then send a short read-only request in isolation; do not guess missing parameters.
  5. Verify text, needed tools, cancellation, timeouts and recovery separately; save results and limits, revert on failure and retain the untested label.
  • Model templates, tool parsing, authentication and field compatibility require testing.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Harness model configuration ↗vLLM official documentation ↗

Codex CLI × Ollama

Official support

Codex CLI documents --oss local-provider mode.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Open the official sources below and record installed versions, model ID and current config; no end-to-end tested version is pinned here.
  2. Codex CLI documents --oss local-provider mode.
  3. Use the source configuration page to select the actual provider and protocol, then send a short read-only request in isolation; do not guess missing parameters.
  4. Verify text, needed tools, cancellation, timeouts and recovery separately; save results and limits, revert on failure and retain the untested label.
  • Requires a suitable local model and endpoint; cloud-service capabilities are not inherited.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Codex advanced configuration ↗Ollama official documentation ↗

Codex CLI × LM Studio

Official support

Codex CLI documents --oss local-provider mode.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Open the official sources below and record installed versions, model ID and current config; no end-to-end tested version is pinned here.
  2. Codex CLI documents --oss local-provider mode.
  3. Use the source configuration page to select the actual provider and protocol, then send a short read-only request in isolation; do not guess missing parameters.
  4. Verify text, needed tools, cancellation, timeouts and recovery separately; save results and limits, revert on failure and retain the untested label.
  • Requires a suitable local model and endpoint; cloud-service capabilities are not inherited.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Codex advanced configuration ↗LM Studio official documentation ↗

Cline × Ollama

Official support

Cline officially lists this local runtime provider.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Open the official sources below and record installed versions, model ID and current config; no end-to-end tested version is pinned here.
  2. Cline officially lists this local runtime provider.
  3. Use the source configuration page to select the actual provider and protocol, then send a short read-only request in isolation; do not guess missing parameters.
  4. Verify text, needed tools, cancellation, timeouts and recovery separately; save results and limits, revert on failure and retain the untested label.
  • A provider entry does not guarantee reliable tool calling for every model.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Cline official documentation ↗Ollama official documentation ↗

Cline × LM Studio

Official support

Cline officially lists this local runtime provider.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Open the official sources below and record installed versions, model ID and current config; no end-to-end tested version is pinned here.
  2. Cline officially lists this local runtime provider.
  3. Use the source configuration page to select the actual provider and protocol, then send a short read-only request in isolation; do not guess missing parameters.
  4. Verify text, needed tools, cancellation, timeouts and recovery separately; save results and limits, revert on failure and retain the untested label.
  • A provider entry does not guarantee reliable tool calling for every model.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Cline official documentation ↗LM Studio official documentation ↗

Home Assistant × Ollama

Official support

Home Assistant’s official Ollama conversation integration optionally enables control.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Prepare Home Assistant and Ollama on a trusted network with a tool-capable model.
  2. Add the Ollama integration using the official UI, set its URL and leave control disabled initially.
  3. After checking conversation, expose one test light and enable experimental control.
  4. Verify on/off, cancellation and invalid requests; keep manual controls and a way to disable AI control.
  • Control is experimental and needs a tool-capable model; begin with low-impact lighting.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Home Assistant official documentation ↗Ollama official documentation ↗

Godot HTTPRequest × Ollama

Partial compatibility

Godot HTTPRequest and Ollama HTTP API can be used to design a custom NPC dialogue prototype.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Prepare an isolated Godot test scene and running local model; pin versions.
  2. Verify model responses with a local HTTP client, then send bounded JSON through HTTPRequest.
  3. Implement timeouts, error handling, cancellation and fixed-dialogue fallback; treat model text as data.
  4. Test normal, disconnected and timed-out cases; add a server boundary protecting credentials before distribution.
  • This is an interface-based design inference, not an official packaged plugin or a site test.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Godot HTTPRequest official documentation ↗Ollama official documentation ↗

Android Debug Bridge × scrcpy

Official support

scrcpy documents ADB connections to authorized Android devices.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Install Platform Tools from Android’s official channel and verify the download.
  2. Enable developer options and USB debugging on your device, connect USB, and approve this computer on the phone.
  3. Use adb devices to confirm device status; resolve unauthorized on the phone without bypassing approval.
  4. Choose an official scrcpy release, start mirroring, and manually test a low-impact action.
  • They provide device communication and screen control, not reliable AI planning.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Android Debug Bridge official documentation ↗scrcpy official documentation ↗

Cline × Model Context Protocol

Official support

Cline documents MCP extension support.

Not tested by BlueFish · Last checked 2026-10-15

Advanced details
  1. Check Cline and server versions, official MCP configuration and server source code.
  2. Inspect startup commands, permissions, authentication and destinations; statically inspect unknown servers without auto-execution.
  3. After user approval, add the server per Cline docs and enable only needed low-impact tools.
  4. Test read-only calls, denied access and cancellation in isolation; record results and retain a removal path.
  • Audit each server separately; MCP support is not a safety certification.
  • Official documentation reviewed; this site has not run this combination or tested the device.
Cline official documentation ↗Model Context Protocol official documentation ↗