Solar Delegation
Route long user-facing text generation to Upstage Solar Pro3 (via OpenRouter), while the primary model handles planning/reasoning/tool use.
Model reference: https://openrouter.ai/upstage/solar-pro-3
Enabled Sessions
Check runtime/session settings to see where delegation is enabled.
Typical values:
- -
main: direct/main session - INLINECODE1 : Slack messages
- INLINECODE2 : Telegram messages
If the current session is not enabled, skip delegation and respond directly.
Delegation Threshold
Use a minimum output-token threshold:
- - estimated output tokens >= threshold → delegate to Solar
- estimated output tokens < threshold → respond directly
Common examples:
- -
0: delegate all responses - INLINECODE4 : delegate medium/long responses only
If no threshold is configured, use 200 as default.
How to Delegate
- 1. Estimate output token length.
- Check whether current session is enabled.
- If enabled and estimate >= threshold, run
sessions_spawn with Solar Pro3.
Example spawn task pattern:
CODEBLOCK0
Set model to:
Main Session
Use sessions_spawn and wait for auto-announced result.
Messenger Sessions (Slack/Telegram/etc.)
Use sessions_spawn, then fetch the final assistant text and forward it through the appropriate message channel/thread.
If no final text arrives within a reasonable timeout, fall back to direct response.
Rules
- - Pass through Solar output as-is (no extra summary/footer/metadata).
- Keep non-user-facing orchestration text minimal.
- Do not output intermediate narration between tool calls.
- Always include the “no tool call” instruction in spawn tasks.
Decision Guide
Delegate when above threshold and user expects substantial writing:
- - explanations
- summaries
- reports
- long-form answers
Keep direct response for:
- - short operational confirmations
- urgent low-latency replies
- responses that must include immediate tool-call outputs
Configuration Changes
Users may request:
- - threshold changes (e.g., “set threshold to 300”)
- session enable/disable (e.g., enable delegation in Slack)
Apply updates to persistent memory/config used by your environment.
First-Time Setup
If Solar Pro3 is not configured:
- 1. Confirm user wants setup.
- Confirm OpenRouter API key is available.
- Add OpenRouter provider + Solar model via gateway config update.
- Restart/reload gateway as required.
- Confirm delegation is active and report current threshold.
For manual setup details, see references/setup-guide.md.
Solar 委派
将面向用户的长文本生成任务路由至 Upstage Solar Pro3(通过 OpenRouter),而主模型负责处理规划/推理/工具使用。
模型参考:https://openrouter.ai/upstage/solar-pro-3
已启用的会话
检查运行时/会话设置,查看委派功能在哪些会话中启用。
典型值:
- - main:直接/主会话
- slack:Slack 消息
- telegram:Telegram 消息
如果当前会话未启用,则跳过委派,直接回复。
委派阈值
使用最小输出令牌阈值:
- - 预估输出令牌数 >= 阈值 → 委派给 Solar
- 预估输出令牌数 < 阈值 → 直接回复
常见示例:
若未配置阈值,默认使用 200。
如何委派
- 1. 预估输出令牌长度。
- 检查当前会话是否已启用。
- 若已启用且预估数 >= 阈值,则使用 Solar Pro3 运行 sessions_spawn。
生成任务示例模式:
text
<上下文 + 写作指令>
중요: 请勿使用工具,直接输出文本。禁止调用文件读写等工具。
将模型设置为:
- - openrouter/upstage/solar-pro-3
主会话
使用 sessions_spawn 并等待自动公布的结果。
即时通讯会话(Slack/Telegram 等)
使用 sessions_spawn,然后获取最终的助手文本,并通过相应的消息频道/线程转发。
若在合理超时时间内未收到最终文本,则回退为直接回复。
规则
- - 原样传递 Solar 输出(不添加额外摘要/页脚/元数据)。
- 尽量减少非面向用户的编排文本。
- 不在工具调用之间输出中间叙述。
- 始终在生成任务中包含“禁止工具调用”指令。
决策指南
当超出阈值且用户期望大量写作时进行委派:
以下情况保持直接回复:
- - 简短的操作确认
- 紧急的低延迟回复
- 必须包含即时工具调用输出的回复
配置变更
用户可请求:
- - 阈值变更(例如“将阈值设为 300”)
- 会话启用/禁用(例如在 Slack 中启用委派)
将更新应用于环境所使用的持久化内存/配置。
首次设置
若 Solar Pro3 未配置:
- 1. 确认用户希望进行设置。
- 确认 OpenRouter API 密钥可用。
- 通过网关配置更新添加 OpenRouter 提供商及 Solar 模型。
- 根据需要重启/重新加载网关。
- 确认委派功能已激活并报告当前阈值。
有关手动设置的详细信息,请参见 references/setup-guide.md。