# DeepSeek Harness · SI-API All 7 SI-API models · one Agent setup Default startup model: deepseek-flash Surface: web System: macos Included models: GLM 5.3 Flash (glm-5.3-flash), DeepSeek V4.1 Flash (deepseek-flash), Kimi K3.1 Pro (kimi-k3.1-pro), Kimi K3.1 Pro HighSpeed (kimi-k3.1-pro-highspeed), GLM 5.3 (glm-5.3), MiniMax M3.1 Flash (minimax-m3.1), Qwen3.8 Max (qwen3.8-max). Templates contain placeholders only. Save your real key locally. Keep unrelated settings and permissions. ## 1. Prepare your tool and key Install DeepSeek Harness from its official page. Create or copy your SI-API key from /keys. Open Local Web → Settings → Models. Add a custom OpenAI-compatible provider named SI-API, ID siapi, base URL https://siapi.ai/v1, and your key. Merge the supplied reasoning.patch.yaml into that provider to add all seven models and their effort controls. ## 2. Save the configuration Download and extract the ZIP. Replace YOUR_SIAPI_API_KEY with your key, then save each file to the path below. Merge these fields into an existing file, preserving other providers, permissions and preferences. Do not replace an existing configuration wholesale. Configuration location: Local Web → Settings → Models ## 3. Start your Agent Start a new session and send a message. The startup model is deepseek-flash. dsh web ## 4. Switch between all models After setup, choose any model below inside your Agent. The same key and configuration work for all seven models; no new download or connection is needed. ### GLM 5.3 Flash · glm-5.3-flash Model picker → SI-API → GLM 5.3 Flash (glm-5.3-flash) Thinking settings: Low, High and Max. Thinking is always enabled; choose Low for less reasoning. The model default is Max. ### DeepSeek V4.1 Flash · deepseek-flash Model picker → SI-API → DeepSeek V4.1 Flash (deepseek-flash) Thinking settings: Off, Low, High and Max. Medium / XHigh map to High; Ultra maps to Max. ### Kimi K3.1 Pro · kimi-k3.1-pro Model picker → SI-API → Kimi K3.1 Pro (kimi-k3.1-pro) Thinking settings: Low, High and Max. Thinking is always enabled; choose Low for less reasoning. The model default is Max. ### Kimi K3.1 Pro HighSpeed · kimi-k3.1-pro-highspeed Model picker → SI-API → Kimi K3.1 Pro HighSpeed (kimi-k3.1-pro-highspeed) Thinking settings: Low, High and Max. Thinking is always enabled; choose Low for less reasoning. The model default is Max. ### GLM 5.3 · glm-5.3 Model picker → SI-API → GLM 5.3 (glm-5.3) Thinking settings: Low, High and Max. Thinking is always enabled; choose Low for less reasoning. The model default is Max. ### MiniMax M3.1 Flash · minimax-m3.1 Model picker → SI-API → MiniMax M3.1 Flash (minimax-m3.1) Thinking settings: Use the model default. This route does not expose custom thinking controls. ### Qwen3.8 Max · qwen3.8-max Model picker → SI-API → Qwen3.8 Max (qwen3.8-max) Thinking settings: Off, Low, Medium and XHigh. High / Max map to XHigh, not additional levels. ## 5. Choose reasoning effort Choose a model in your Agent, then use its effort or variant selector. For DeepSeek V4.1 Flash, try Low and High. Start on DeepSeek V4.1 Flash. Choose Low, send a short file-reading task, then choose High and repeat. Check the current setting after each change. Effort means how much reasoning to request. Kimi K3.1 Pro / HighSpeed, GLM 5.3 and GLM 5.3 Flash offer Low / High / Max with thinking always enabled. Choose Low for shorter reasoning; there is no Off option. Their model default is Max, but a client setting can override it. DeepSeek and Qwen have their own intensity levels listed below. MiniMax M3.1 Flash uses its default reasoning behavior. MiniMax M3 settings apply to a different model. Model discovery or manual entry may provide names without reasoning metadata. If the Effort menu is missing, stop the app and merge the models below into the existing siapi provider in the active profile's cordis.patch.yml. For dsh web, this is $DSH_HOME/profiles/web/cordis.patch.yml; use Open configuration file in the Settings header to locate it when available. Desktop has its own profile: locate the profile used by that app instead of editing the web profile. Preserve the existing plugin identifier, API key reference, URL and other providers. Current profiles identify this plugin as id: llm-pi-ai; older profiles can use name: '@deepseek-ai/dsh-llm-pi-ai'. Update existing model entries by id instead of duplicating them. Restart, select DeepSeek V4.1 Flash, then open the model picker's Effort menu and choose Low or High. The picker saves the selection as the default for new sessions. This download is a capability patch, not a connection configuration or a thinking default. Before the first request on every model, select a legal effort explicitly: High for DeepSeek, Max for Kimi and GLM, XHigh for Qwen. An unset selection can leave thinking Off on some models. After switching models, inspect the effort selection again. ### Merge model capabilities into the active profile ```yaml - id: llm-pi-ai config: providers: siapi: compat: supportsDeveloperRole: false models: - id: glm-5.3-flash reasoningEfforts: low: low high: high max: max - id: deepseek-flash reasoningEfforts: off: none low: low high: high max: max - id: kimi-k3.1-pro reasoningEfforts: low: low high: high max: max - id: kimi-k3.1-pro-highspeed reasoningEfforts: low: low high: high max: max - id: glm-5.3 reasoningEfforts: low: low high: high max: max - id: minimax-m3.1 reasoningEfforts: false - id: qwen3.8-max reasoningEfforts: off: none low: low medium: medium xhigh: xhigh ``` - Off maps explicitly to none. An empty off value only omits a parameter and can leave the server's default thinking enabled. Keep the supplied mappings. Expected: On DeepSeek V4.1 Flash, the current setting changes from Low to High and back to Low before your next message. Other models follow the controls listed below. Harness profiles and reasoningEfforts: https://github.com/deepseek-ai/deepseek-harness/blob/master/docs/user/guide/providers.md#reasoning-effort ## 6. Check the model and run your first task Open a project containing README.md. Allow the tool to read it, then send the prompt below. Confirm that it reads the file and returns a summary. To check another model, switch using the table above and repeat this task. Check the active SI-API model in the model picker beside the composer. Use the file-reading tool to read README.md in this project and summarize its purpose in one sentence. Do not modify any files. ## provider-reference.json Save to: Local Web → Settings → Models Use this reference to create provider siapi and enter your key in the local credential field. Merge reasoning.patch.yaml below to add all seven models and their effort controls together. ## reasoning.patch.yaml Save to: The active profile’s cordis.patch.yml Merge the model entries into the existing llm-pi-ai provider siapi. Preserve its connection and credential fields; do not add a duplicate plugin. Keep the existing plugin identifier (older profiles use name: '@deepseek-ai/dsh-llm-pi-ai'). This patch declares effort choices; select a supported Effort in the model picker before the first request. ## Notes Desktop and local Web can use different profiles. Edit the active profile shown by that app; Web normally uses $DSH_HOME/profiles/web/cordis.patch.yml. If the key is supplied through SIAPI_API_KEY, the process running Harness must inherit it. Otherwise use its local credential store. The reference file contains no credential. Off, Low, High and Max. Medium / XHigh map to High; Ultra maps to Max. Reference: https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/llm/llm-pi-ai/README.md Model configuration: https://siapi.ai/docs/models#configuration Agent compatibility: https://siapi.ai/docs/compatibility