* feat(agents): support `thinkingDefault: "adaptive"` for Anthropic models
Anthropic's Opus 4.6 and Sonnet 4.6 support adaptive thinking where the
model dynamically decides when and how much to think. This is now
Anthropic's recommended mode and `budget_tokens` is deprecated on these
models.
Add "adaptive" as a valid thinking level:
- Config: `agents.defaults.thinkingDefault: "adaptive"`
- CLI: `/think adaptive` or `/think auto`
- Pi SDK mapping: "adaptive" → "medium" effort at the pi-agent-core
layer, which the Anthropic provider translates to
`thinking.type: "adaptive"` with `output_config.effort: "medium"`
- Provider fallbacks: OpenRouter and Google map "adaptive" to their
respective "medium" equivalents
Closes#30880
Made-with: Cursor
* style(changelog): format changelog with oxfmt
* test(types): fix strict typing in runtime/plugin-context tests
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
GitHub Copilot API tokens expire after ~30 minutes. When OpenClaw spawns
a long-running subagent using Copilot as the provider, the token would
expire mid-session with no recovery mechanism, causing 401 auth errors.
This commit adds:
- Periodic token refresh scheduled 5 minutes before expiry
- Auth error detection with automatic token refresh and single retry
- Proper timer cleanup on session shutdown to prevent leaks
The implementation uses a per-attempt retry flag to ensure each auth
error can trigger one refresh+retry cycle without creating infinite
retry loops.
🤖 AI-assisted: This fix was developed with GitHub Copilot CLI assistance.
Testing: Fully tested with 3 new unit tests covering auth retry, retry
reset, and timer cleanup scenarios. All 11 auth rotation tests pass.
Fix tool-call lookup failures when models emit whitespace-padded names by normalizing
both transcript history and live streamed embedded-runner tool calls before dispatch.
Co-authored-by: wangchunyue <80630709+openperf@users.noreply.github.com>
Co-authored-by: Sid <sidqin0410@gmail.com>
Co-authored-by: Philipp Spiess <hello@philippspiess.com>
* fix(agents): add "google" provider to isReasoningTagProvider to prevent reasoning leak
The gemini-api-key auth flow creates a profile with provider "google"
(e.g. google/gemini-3-pro-preview), but isReasoningTagProvider only
matched "google-gemini-cli" (OAuth) and "google-generative-ai". As a
result:
- reasoningTagHint was false → system prompt omitted <think>/<final>
formatting instructions
- enforceFinalTag was false → <final> tag filtering was skipped
Raw <think> reasoning output was delivered to the end user.
Fix: add the bare "google" provider string to the match list and cover
it with two new test cases (exact match + case-insensitive).
Fixes#26551
* fix(agents): add forward-compat fallback for google-gemini-cli gemini-3.1-pro/flash-preview
gemini-3.1-pro-preview and gemini-3.1-flash-preview are not yet present in
pi-ai's built-in google-gemini-cli model catalog (only gemini-3-pro-preview
and gemini-3-flash-preview are registered). When users configure these models
they get "Unknown model" errors even though Gemini CLI OAuth supports them.
The codebase already has isGemini31Model() in extra-params.ts, which proves
intent to support these models. Add a resolveGoogleGeminiCli31ForwardCompatModel
entry to resolveForwardCompatModel following the same clone-template pattern
used for zai/glm-5 and anthropic 4.6 models.
- gemini-3.1-pro-* clones gemini-3-pro-preview (with reasoning: true)
- gemini-3.1-flash-* clones gemini-3-flash-preview (with reasoning: true)
Also add test helpers and three test cases to model.forward-compat.test.ts.
Fixes#26524
* Changelog: credit Google Gemini provider fallback fixes
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
Azure OpenAI endpoints were not recognized by shouldForceResponsesStore(),
causing store=false to be sent with all Azure Responses API requests.
This broke multi-turn conversations because previous_response_id referenced
responses that Azure never stored.
Add "azure-openai-responses" to the provider whitelist and
*.openai.azure.com to the URL check in isDirectOpenAIBaseUrl().
Fixes#27497
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
(cherry picked from commit 185f3814e94cf50bf04838994af1944389a4b6c0)
When reasoningLevel is 'on', reasoning content was being sent as a
visible message to WhatsApp and other non-Telegram channels via two
paths:
1. Block reply: emitted via onBlockReply in handleMessageEnd
2. Final payloads: added to replyItems in buildEmbeddedRunPayloads
Telegram has its own dispatch path (bot-message-dispatch.ts) that
splits reasoning into a dedicated lane and handles suppression.
The generic dispatch-from-config.ts path used by WhatsApp, web, etc.
had no such filtering.
Fix:
- Add isReasoning?: boolean flag to ReplyPayload
- Tag reasoning payloads at both emission points
- Filter isReasoning payloads in dispatch-from-config.ts for both
block reply and final reply paths
Telegram is unaffected: it uses its own deliver callback that detects
reasoning via the 'Reasoning:\n' prefix and routes to a separate lane.
Fixes#24954
The 'auto' model on OpenRouter dynamically routes to any underlying model
OpenRouter selects, including reasoning-required endpoints. Previously,
OpenClaw would unconditionally inject `reasoning.effort: "none"` into
every request when the thinking level was "off", which causes a 400 error
on models where reasoning is mandatory and cannot be disabled.
Root cause:
- openrouter/auto has reasoning: false in the built-in catalog
- With thinking level "off", createOpenRouterWrapper injects
`reasoning: { effort: "none" }` via mapThinkingLevelToOpenRouterReasoningEffort
- For any OpenRouter-routed model that requires reasoning this results in:
"400 Reasoning is mandatory for this endpoint and cannot be disabled"
- The reasoning: false is then persisted back to models.json on every
ensureOpenClawModelsJson call, so manually removing it has no lasting effect
Fix:
- In applyExtraParamsToAgent, when provider is "openrouter" and the model
id is "auto", pass undefined as thinkingLevel to createOpenRouterWrapper
so no reasoning.effort is injected at all, letting OpenRouter's upstream
model handle it natively
- Add an explanatory comment in buildOpenrouterProvider clarifying that the
reasoning: false catalog value does NOT cause effort injection for "auto"
Users who need explicit reasoning control should target a specific model
id (e.g. openrouter/deepseek/deepseek-r1) rather than the auto router.
Fixes#24851
(cherry picked from commit aa554397980972d917dece09ab03c4cc15f5d100)
"off" is a truthy string, so the existing guard `if (thinkingLevel && ...)`
was always entering the injection block and sending `reasoning: { effort: "none" }`
to every OpenRouter request — even when thinking wasn't enabled. Models that
require reasoning (e.g. deepseek/deepseek-r1) reject this with:
400 Reasoning is mandatory for this endpoint and cannot be disabled.
Fix: skip the reasoning injection entirely when thinkingLevel is "off".
The reasoning_effort flat-field cleanup still runs. Omitting the reasoning
field lets each model use its own default behavior.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
sessions_send timeout/error results were being surfaced as raw warning
messages in Telegram chats because the tool is classified as mutating,
which forces error warnings to always be shown. However, sessions_send
failures are transient inter-session communication issues where the
message may still have been delivered, so they should not leak to users.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Isolated cron sessions (agentTurn) were grouped with subagent sessions
under the "minimal" prompt mode, which causes buildSkillsSection to
return an empty array. This meant <available_skills> was never included
in the system prompt for isolated cron runs.
Subagent sessions legitimately need minimal prompts (reduced context),
but isolated cron sessions are full agent turns that should have access
to all configured skills, matching the behavior of normal chat sessions
and non-isolated cron runs.
Remove isCronSessionKey from the minimal prompt condition so only
subagent sessions use "minimal" mode.
Fixes openclaw#24888
Co-authored-by: Cursor <cursoragent@cursor.com>
* feat: Add Kilo Gateway provider
Add support for Kilo Gateway as a model provider, similar to OpenRouter.
Kilo Gateway provides a unified API that routes requests to many models
behind a single endpoint and API key.
Changes:
- Add kilocode provider option to auth-choice and onboarding flows
- Add KILOCODE_API_KEY environment variable support
- Add kilocode/ model prefix handling in model-auth and extra-params
- Add provider documentation in docs/providers/kilocode.md
- Update model-providers.md with Kilo Gateway section
- Add design doc for the integration
* kilocode: add provider tests and normalize onboard auth-choice registration
* kilocode: register in resolveImplicitProviders so models appear in provider filter
* kilocode: update base URL from /api/openrouter/ to /api/gateway/
* docs: fix formatting in kilocode docs
* fix: address PR review — remove kilocode from cacheRetention, fix stale model refs and CLI name in docs, fix TS2742
* docs: fix stale refs in design doc — Moltbot to OpenClaw, MoltbotConfig to OpenClawConfig, remove extra-params section, fix doc path
* fix: use resolveAgentModelPrimaryValue for AgentModelConfig union type
---------
Co-authored-by: Mark IJbema <mark@kilocode.ai>
* Telegram: soft-fail reactions and fallback to inbound message id
* Telegram: soft-fail missing reaction message id
* Update CHANGELOG.md
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
