Repository navigation
[DOCS] Document chat.suggestions next-step suggestion configuration #1005
Description
Summary
PR inference-gateway/cli#1554 adds an opt-in one-line next-step suggestion, rendered as dim ghost text in the empty chat input after an assistant turn completes. The CLI repo's own reference page was updated in the same PR, but the standalone docs site needs to cover the feature for users.
Affected pages
- Configuration reference page for the chat section (mirrors inference-gateway/cli docs/configuration-reference.md)
- Any page enumerating INFER_* environment overrides (new INFER_CHAT_SUGGESTIONS_* variables and the INFER_PROMPTS_CHAT_SUGGESTIONS_SYSTEM_PROMPT prompt override)
- TUI feature overview wherever the input placeholder and history-suggestion behavior is described
Acceptance Criteria
- Document chat.suggestions.enabled, chat.suggestions.model, chat.suggestions.context_messages, chat.suggestions.max_tokens and chat.suggestions.timeout, including defaults (false, empty falls back to agent.model, 4, 64, 10 seconds)
- Document the Tab / Right Arrow accept-to-input behavior and the Esc, typing and new-turn dismissal
- List the INFER_CHAT_SUGGESTIONS_* environment overrides next to the other chat overrides
- Mention the silent-fail behavior and that the suggestion costs one extra small LLM call per completed turn
- Use plain hyphens only, no en or em dashes
Activity
feat(chat): one-line next-step suggestion as input ghost text (#1554)
edenreich commentedon Oct 12, 2026
@claude can you document this?
inference-gateway-maintainer commentedon Oct 12, 2026
Claude finished @edenreich's task in 2m 31s —— View job •
docs/issue-1005-20261012-1732Documented
chat.suggestionsdocs/configuration-reference.md, localcli.mdchat sections)chat.suggestionskeys, defaults and env overrideslint:md,format:check,buildall pass)All changes landed in
cli.md:#### Next-Step Suggestionssection under Chat Interface Features (right after the prompt-history search section): enabling it, the two-second quiet-window trigger, the cost note, a keybinding table for accept and dismissal, the silent-fail behavior, a YAML example and a key table with defaults.INFER_CHAT_SUGGESTIONS_*plusINFER_PROMPTS_CHAT_SUGGESTIONS_SYSTEM_PROMPTadded to the Essential Environment Variables block, and a one-line mention in the chat capabilities list.One deviation worth flagging: the issue lists
max_tokensdefault as64, but the CLI repo's currentdocs/configuration-reference.mdsays512. I documented512as the canonical upstream value - tell me if the issue is the right one and I will flip it.Kept it to
cli.mdrather than touchingconfiguration.md, whose chat section is generated from the gateway OpenAPI schema and does not carry CLI keys.Create a PR - already opened as #1006.
· branch
inference-gateway-releaser commentedon Oct 19, 2026
🎉 This issue has been resolved in version 0.9.22 🎉
The release is available on GitHub release
Your semantic-release bot 📦🚀