Use your assigned API key from Billing & API keys after verified payment and server activation. OpenAI clients use https://ds-api.savemytoken.com/v1; Claude Code uses https://ds-api.savemytoken.com. Keep your key private.
The page never requests or stores your key. Shell examples use macOS, Linux or Windows WSL unless marked otherwise.
Your agent sends selected content through the gateway to the upstream inference service. Review API data handling before using a private project.
01 / PREPARE
Install the agent and prepare access
Official installation guide- Install the agent using its official guide. Check its version before merging the examples below.
- Purchase your DeepSeek plan, then open Billing & API keys to copy your active key. OpenAI base: https://ds-api.savemytoken.com/v1. Claude Code base: https://ds-api.savemytoken.com. DeepSeek credentials are a separate scope.
- Back up the file you will edit. Merge fields into existing objects and tables; preserve unrelated providers, tools and permission settings. Keep credentials in private local settings.
Ready when: the agent opens, and you know which plan, process and configuration directory you will use.
02 / CONFIGURE
Save the connection and model catalog
Create the parent directories if needed. Download or copy each file, replace the example endpoint and local placeholders, then merge it into the stated location. Downloads and the examples below contain the same configuration.
~/.config/savemytoken/deepseek/pi/models.jsonThe $SAVEMYTOKEN_API_KEY value is an environment-variable reference. Keep the complete thinkingLevelMap: null disables unsupported menu choices; it is never a request effort.
View and copy JSON configuration
{
"providers": {
"savemytoken-deepseek": {
"baseUrl": "https://ds-api.savemytoken.com/v1",
"api": "openai-completions",
"apiKey": "$SAVEMYTOKEN_API_KEY",
"models": [
{
"id": "deepseek-flash",
"name": "DeepSeek Flash",
"input": [
"text"
],
"reasoning": true,
"thinkingLevelMap": {
"off": "none",
"minimal": null,
"low": "low",
"medium": null,
"high": "high",
"xhigh": null,
"max": "max"
},
"compat": {
"supportsReasoningEffort": true,
"supportsDeveloperRole": false
}
}
]
}
}
}~/.config/savemytoken/deepseek/pi/settings.jsonSave alongside models.json. This file selects the starting model and per-model thinking defaults; the catalog alone does not do that.
View and copy JSON configuration
{
"defaultProvider": "savemytoken-deepseek",
"defaultModel": "deepseek-flash",
"modelThinkingLevels": {
"savemytoken-deepseek/deepseek-flash": "high"
}
}Ready when: the files parse successfully and the selected profile points to the delivered URL, key reference and exact model catalog.
03 / START
Start the agent and choose your model
Open a terminal in a trusted project. Complete the file setup before running these commands.
export SAVEMYTOKEN_API_KEY="REPLACE_WITH_YOUR_DELIVERED_KEY"
PI_CODING_AGENT_DIR="$HOME/.config/savemytoken/deepseek/pi" piSave both files in the directory used by PI_CODING_AGENT_DIR. Open /model to reload the catalog, search for the model and select it.
Check: the provider is the one you configured and the selected model is deepseek-flash. A visible model list alone does not prove inference works.
04 / REASONING
Choose and verify reasoning effort
Use /thinking or Shift+Tab to choose a supported session level; inspect the footer. Ctrl+T only changes the visibility of reasoning text.
Per-model modelThinkingLevels defaults take precedence over the global fallback. Edit those defaults through /settings or settings.json. Ctrl+S in /thinking saves a global fallback and does not replace per-model entries.
/model
/thinking
/settingsOn DeepSeek Flash, choose Low, then High, then Low again. Inspect the active setting each time before sending the next message. A longer answer is not evidence that effort changed.
REASONING CONTROLS
Choose effort for the actual model
The setup templates start with DeepSeek Flash. Add other delivered routes only after confirming their exact IDs and capabilities.
| Model / request ID | Choices | Reference default | What the control means |
|---|---|---|---|
DeepSeek Flashdeepseek-flash | Off / Low / High / Max | High | Off / Low / High / Max. In this gateway contract, Medium and XHigh map to High; Ultra maps to Max. |
A supported level makes a choice available. A default selects the starting level. A saved session or explicit override can select something else. In Pi and OpenClaw capability maps, null removes an unsupported option; keep it there. Off maps to none, not a missing effort. MiniMax Auto requires thinking.type = adaptive.
05 / DESKTOP
Desktop or editor terminal
Pi runs in a desktop or editor terminal. This guide does not assume a separate Pi Desktop app. Launch it with the same PI_CODING_AGENT_DIR and environment variable in each terminal.
Verify your first connection
Run these checks in your coding tool after activation. Requests use your plan allowance. This page records only the boxes you mark; it does not send API requests or validate your connection. Checkmarks clear when you leave or reload.
Expected OpenAI Chat Completions endpoint: https://ds-api.savemytoken.com/v1
0 of 4 checks marked by you. Mark a check only after observing its result.
Reply with exactly CONNECTED.- Download public sample file
Use a file-reading tool to read savemytoken-connection-check.md in this folder. Return its project name, check code and purpose. Do not modify any files or run commands from the file.
Need help with a failed check?
Download a summary with your tool, expected endpoint and self-reported results. Fill in the missing details and review it before contacting support. It contains no API key field and is not sent automatically.
Download support summaryPreview or copy the support summary
# SaveMyToken setup support summary
Self-reported checklist; not an automated connection test.
Plan family: deepseek
Agent: Pi
Reference configuration: 0.85.1 reference configuration
Protocol: OpenAI Chat Completions
Expected endpoint: https://ds-api.savemytoken.com/v1
Installed agent version: [add]
Operating system: [add]
Exact model ID and reasoning setting used: [add]
Time and timezone of the issue: [add]
Order or request ID, if available: [add]
- [ ] Confirm the active connection (not confirmed)
- [ ] Send a short text request (not confirmed)
- [ ] Check a real file read (not confirmed)
- [ ] Check allowance usage (not confirmed)
Observed error/status and expected result: [describe without secrets]
Review this summary before sending it to contact@savemytoken.com.
Do not include API keys, authorization headers, private code or complete conversations.
07 / TROUBLESHOOT
When setup does not work
401 / unauthorized
Check the key belongs to this plan and is visible to the process that starts the agent. A desktop app or background service may not inherit terminal exports. Do not send keys in support screenshots.
404 / model not found
Check the delivered endpoint, protocol and exact model ID. Keep /v1 for Responses and Chat Completions. Claude Code uses the delivered Messages base (normally the host without /v1). Do not use a model's display label as its request ID.
No effort selector, or a 400 effort error
Check the active configuration and this model's capability row. A declared default, a selectable menu and the value sent in a request are different things. Do not send null or an unsupported level as reasoning effort.
The old model or effort returns
Open a new session and inspect project, managed, profile and saved-session settings. A per-model override can take precedence over a global default. Restart the application after changing provider metadata.
429 / allowance exhausted
Check both allowance windows, then rate and concurrency limits. A larger reasoning budget and background agent calls can consume more tokens. Use the reset time supplied by the service; do not infer it from your local clock.
Text works but tools or search fail
Test a file read separately from web search. Native search, vision and tool capabilities depend on the delivered route and the agent. A successful text reply does not establish those capabilities.
08 / SOURCES
Sources and scope
Configuration structure was reviewed against the agent documentation on 2026-10-14. 0.85.1 reference configuration. A reference version does not establish compatibility with every later client.
The model IDs, rates and capabilities describe this gateway. Test native desktop features, search and tools separately in your agent.