memory optimization

OpenRouter provider fails with "401 Missing Authentication header" despite valid API key

OpenClaw's OpenRouter provider is not sending the Authorization: Bearer <token> header, causing all requests to fail with 401 authentication errors. Environment • OpenClaw version: 2026.3.13 (61d171a) • OS: Linux (Debian-based) • Node version: v22.22.1 Configuration // ~/.openclaw/openclaw.json "models": { "providers": { "openrouter": { "baseUrl": "https://openrouter.ai/api/v1", "api": "openai-completions", "models": [...] } } }, "auth": { "profiles": { "openrouter:default": { "provider": "openrouter", "mode": "api_key" } } }

Bug in Release Notes (v2026.3.31): 'openclaw flows' is actually 'openclaw tasks'

[Bug]: [Bug]: Requests abort after ~1 minute with Request timed out before a response was generated despite agents.defaults.timeoutSeconds set to 18 minutes

OpenClaw aborts requests after about 1 minute with Request timed out before a response was generated even though agents.defaults.timeoutSeconds is set to 18 minutes. <img width="1050" height="537" alt="Image" src="https://github.com/user-attachments/assets/963947c8-09cf-4902-88f4-76ea46eb40e4" />

[Bug]: The hit rate of prefix caching in Qwen3.5 35BA3B is very low, always less than 0.1%

[Bug] Telegram getUpdates 409 conflict when multiple bot accounts poll from same gateway

When running a single gateway with multiple Telegram bot accounts (8 in our case), the gateway produces persistent `409 Conflict` errors on `getUpdates` — even though every account has a **unique bot token**.

ROCm error: invalid device function in ggml_cuda_mul_mat_q on RX 6750 GRE (gfx1031) with ollama-for-amd patch on Windows

[Bug]: 2026.3.28 bundled feishu channelConfigs schema overrides openclaw-lark plugin schema even when bundled plugin is disabled

When upgrading from 2026.3.24 to 2026.3.28, Gateway refuses to start with: ``` channels.feishu: invalid config: must NOT have additional properties channels.feishu.accounts.jeeves: invalid config: must NOT have additional properties channels.feishu.accounts.ivy: invalid config: must NOT have additional properties channels.feishu.accounts.crichton: invalid config: must NOT have additional properties channels.feishu.accounts.marple: invalid config: must NOT have additional properties ``` The external plugin `@larksuite/openclaw-lark` (2026.3.26) registers channel id `feishu` with a full config schema that includes keys like `replyMode`, `footer`, `botName`, `blockStreaming`, `markdown`, `dedup`, `threadSession`, and `uat`. However, the bundled `@openclaw/feishu` plugin also registers channel id `feishu` with a `channelConfigs` schema (new in 2026.3.28) that does **not** include these keys and sets `additionalProperties: false`. The bundled plugin is explicitly disabled (`plugins.entries.feishu.enabled: false`), but its schema still takes priority during config validation.

[Bug]: Duplicated number omitted in TUI

When Openclaw uses bc for calculation, the TUI output will omit duplicate number, but get normal when reopen the TUI

[Bug] Sandbox mode: DingTalk sendMedia fails for workspace files

[Gemma 4] mm_token_type_ids required for text-only fine-tuning - should default to zeros

[Bug]: WebSocket Gateway Probe Timeout on Loopback

Running openclaw gateway probe returns "Connect: failed - timeout" even though the gateway is fully functional (HTTP works, WebSocket connections work). Environment: OpenClaw: 2026.3.13 OS: Linux 6.12.62+rpt-rpi-2712 (arm64) - Raspberry Pi 5 Gateway: loopback (127.0.0.1:18789) Steps to reproduce:

doctor --fix leaves legacy notify:true cron field unresolved when cron.webhook is unset

`openclaw doctor --fix` does not fully migrate legacy cron metadata when a job still has top-level `notify: true` and `cron.webhook` is unset. Observed behavior: - Doctor reports legacy cron storage and warns about `notify: true` fallback. - Running `openclaw doctor --fix` prints `Cron store normalized at ~/.openclaw/cron/jobs.json.` - But the same legacy warning persists on subsequent doctor runs until the user manually edits `jobs.json` and removes `notify`. This creates a confusing loop where doctor claims normalization succeeded, but legacy state remains.

quwen3-next:80b failed to load 500 server error on ollama v0.17.7

[Bug]: Local model support broken by three compounding failures in model validation, provider routing, and error classification

Feature: per-model Ollama API options in config (e.g. think: false for Qwen 3.5)

[Bug]: Malformed agent session key when spawning subagent via sessions_spawn

Control UI on local loopback (127.0.0.1) still requires pairing in 2026.3.13, despite docs saying local connections auto-approve

Control UI on local loopback (`127.0.0.1`) is still requiring pairing in `2026.3.13`, even though the docs say local connections are auto-approved.

[Feature Request] Add agent:error event hook support for auto-recovery mechanisms

TTS tool on Telegram renders as audio file instead of native voice memo (sendAudio vs sendVoice)

The built-in `tts` tool on Telegram delivers voice audio using `sendAudio` (or `sendDocument`) instead of `sendVoice`. This causes voice memos to render as generic audio files with clunky metadata (filename, "Unknown artist", music note icon) instead of native Telegram voice memos with a clean waveform UI.

[Bug]: openclaw update fails with preflight-no-good-commit; exec auto-allow/approval flow inconsistent on git install

I hit two problems on a Linux git install of OpenClaw: 1. `openclaw update` failed with `preflight-no-good-commit`. 2. Host exec works, but auto-exec / approval behavior appears inconsistent: - harmless commands were blocked with `Blocked by allowlist. Need approval to proceed.` - no usable approval popup/UI appeared - `/approve ...` failed with `unknown or expired approval id` - manual allowlisting of binaries made the same commands succeed This suggests: - the updater is hiding the actionable dependency-install reason behind a generic preflight error - the exec engine itself works, but the approval / allowlist / auto-exec flow is broken or inconsistent