- Remove vendor-specific preset model lists for custom embedding/vision
providers; users now type the model id manually (lists vary per vendor).
- Strip third-party vendor names from comments.
- Fix AttributeError in _set_vision/_set_embedding: the provider constants
live on ModelsHandler, not ConfigHandler, so selecting a built-in
provider crashed. Also replace the fragile [:-1] slice with an explicit
non-custom filter.
Co-authored-by: Cursor <cursoragent@cursor.com>
1. Embedding model: support custom provider
- Add "custom" entry to EMBEDDING_VENDORS with supports_dim_param=False
- Parse custom:<id> credentials and model fallback in agent_initializer
- Expand custom_providers as custom:<id> entries in Web UI dropdown
2. Vision model: support custom provider
- Add custom:<id> routing in _route_by_provider_id
- Add _build_custom_provider reading credentials from custom_providers
- Expand custom_providers in Web UI dropdown, add validation in _set_vision
3. Fix memory_get Windows path validation bug!
- str.startswith(path+'/') always False on Windows due to backslashes, So All Users can not use "memory_get" tool in Windows.
- Use os.path.realpath + os.sep, consistent with MemoryService
4. Fix historical needsModel:false bug preventing embedding provider switch
- Change embedding needsModel to true in console.js
- Support custom:<id> resolution in cow_cli /memory status, also for adding custom provider support
Closes#2908, Closes#2880
web_fetch fetched any http/https URL a model emitted, checking only the
scheme. It performed requests.get(..., allow_redirects=True) with no
hostname resolution check and no private/loopback/link-local/cloud-metadata
filtering, and never re-validated redirect targets. A model (including one
under prompt injection) could make CowAgent fetch 127.0.0.1, RFC1918,
169.254.169.254 or other internal endpoints and return their bodies into
the conversation; a public URL could also 302-bounce into a private target.
The repo already shipped an SSRF validator for the vision tool
(Vision._validate_url_safe). Extract that logic into a shared helper
(agent/tools/utils/url_safety.py) and reuse it:
- execute() now validates the URL before dispatching to either fetch path.
- A new _safe_get() helper disables auto-redirect and follows redirects
manually, re-validating every hop so a public URL cannot bounce into an
internal address. Both the webpage and document fetch paths use it.
- Vision._validate_url_safe now delegates to the shared helper (public API
unchanged), so both URL-consuming tools share one guard.
Stdlib only (ipaddress, socket, urllib.parse); no new dependency. Adds
tests/test_security_ssrf_web_fetch.py covering loopback, cloud-metadata,
RFC1918 and a public->loopback redirect.
Sink: agent/tools/web_fetch/web_fetch.py (_fetch_webpage / _fetch_document).
Signed-off-by: christop <825583681@qq.com>
- Add Kimi kimi-k2.7-code (default), kimi-k2.7-code-highspeed, and GLM glm-5.2 (default)
- Fix 400 error when disabling thinking on kimi-k2.7-code; omit the thinking param for this series since it only accepts type=enabled
- Update README, docs (zh/en/ja), install scripts, and Web console model dropdown
Replace the boolean wecom_bot_callback with a wecom_bot_mode field
("websocket" | "webhook"), consistent with the Feishu channel's
feishu_event_mode. Update startup() and send() to branch on the mode.
Also fix a temp-file leak in _load_image_base64: downloads, format
conversions and compressions wrote to /tmp but were never removed. Track
only the temp files created here and delete them in a finally block,
leaving the caller's original local file untouched.
Co-authored-by: Cursor <cursoragent@cursor.com>
Remove the callback-mode fields (Callback Mode / Token / EncodingAESKey /
Port) from the web console channel form; these are rarely changed and are
now configured via config.json only. The console keeps Bot ID / Secret for
the long-connection setup.
Serve the callback HTTP server on a fixed path (/wecombot) instead of any
path (/.*), so unrelated requests 404 rather than being processed as
signature-failing WeCom callbacks. The bot's receive-message URL must point
at http(s)://host:<port>/wecombot.
Co-authored-by: Cursor <cursoragent@cursor.com>
The 2MB cap (matching the long-connection upload path) does not work for
the callback path: there the whole image is base64-embedded in an
AES-encrypted body returned on every poll. A ~1.5MB image (base64 ~2.1MB,
encrypted ~2.8MB) makes WeCom reject the finish packet and poll forever,
which also surfaces as a truncated text bubble and WeCom's own timeout
error. Cap well below that at 512KB so the finish packet is accepted.
Co-authored-by: Cursor <cursoragent@cursor.com>
Defer the callback stream finish after a text reply so a trailing
image-with-caption send (text first, image 0.3s later) can merge in
instead of closing the stream prematurely. Raise the callback inline
image cap from 512KB to 2MB to match the long-connection upload path.
Co-authored-by: Cursor <cursoragent@cursor.com>
In callback mode the image is base64-embedded in the stream finish reply and
the whole response is AES-encrypted and returned on every poll. A multi-MB
body is rejected/times out on WeCom's side, leaving the "···" bubble spinning
and the image never shown.
- Compress callback images to <=512KB (JPEG, resize if needed) instead of the
10MB the protocol nominally allows
- Fall back to the original image if compression fails, and log the final
base64 payload size for diagnosis
Co-authored-by: Cursor <cursoragent@cursor.com>
The streaming "···" bubble should only stop when the task actually completes
or the user cancels — not on an arbitrary timeout that would also steal the
response_url fallback's chance to deliver a late answer.
- On agent_cancelled, finalize the stream (finish=true, "🛑 已中止" if empty)
and schedule the response_url fallback so the bubble clears immediately when
a run is cancelled, even past the poll window
- Do not force-finish a still-running stream on a timer; let it keep spinning
until completion or cancel. Answers that finish after WeCom's ~6min poll
window are delivered via response_url instead
Co-authored-by: Cursor <cursoragent@cursor.com>
- Add toggle, update and delete APIs for scheduled tasks
- Add task edit modal with schedule/action updates in web console (PS: In edit box, now I Prevent channel type changes during editing (weixin token bound to session) )
- Add enable/disable switch with visual feedback in task cards
- Sort task list by enabled status first, then by next_run_at
Closes#2882
- Same title overrides built-in role; different title appends as new
- roles/ directory is optional — no impact when absent
Users can now add custom roles by simply dropping a .json file
into the roles/ directory and restarting. No config changes needed.
Handle agent replies that finish after WeCom stops polling the passive
stream (the poll window is ~6min from the user's message).
- Capture response_url from the message callback and, when a reply is
finalized but no poll picks it up within a short grace period, push it as
a one-shot active markdown reply (valid 1h, single use)
- Guard against double delivery via delivered/url_sent flags
- Embed public image URLs in the active markdown; note when a local image
can't be delivered post-timeout
- Append (instead of discarding) the unsupported-type notice for
file/voice/video replies so streamed text is preserved
- Quiet the per-poll debug log and log stream completion with content size
Co-authored-by: Cursor <cursoragent@cursor.com>
- Merge the separate custom-providers section into the unified provider
grid; "Custom" in the add-provider picker now acts as an add-new action
(trailing + mark) and opens the dedicated modal, supporting multiple
OpenAI-compatible endpoints
- Simplify the custom provider modal: drop the default-model field, add
an inline delete button, align colors with the theme
- Keep the legacy single "custom" card visible (models page, chat
dropdown and legacy config page) while custom_api_key/custom_api_base
is still in use, so existing single-provider setups don't disappear
- Unify user-facing wording from "vendor" to "provider" in UI and docs
- Restructure custom provider docs (zh/en/ja) around Web console and
config file usage
Support receiving WeCom smart-bot messages via encrypted HTTP callback in
addition to the existing WebSocket long connection. Disabled by default
(wecom_bot_callback=false).
- Add wecom_bot_callback / wecom_bot_token / wecom_bot_encoding_aes_key /
wecom_bot_port config keys
- Add WXBizJsonMsgCrypt-based crypto module for URL verification, callback
decryption and passive-reply encryption (receive_id empty for internal bots)
- Reply asynchronously via the official stream-refresh polling: register a
stream id on first reply, accumulate agent output into per-stream state, and
serve the latest content (text + image) on each poll until finish
- Fall back to EncodingAESKey for media decryption when callback bodies carry
no per-message aeskey
- Degrade unsupported passive replies (file/voice/video) to a text notice
- Expose the new fields in the Web console channel config
Co-authored-by: Cursor <cursoragent@cursor.com>
_set_chat rejected the expanded "custom:<id>" ids with "unknown
provider", so switching to a custom provider was only possible from
the custom providers section. Now the chat card and the custom section
behave consistently: _set_chat validates the id against
custom_providers, falls back to the provider's default model when none
is picked, and _chat_capability expands the dropdown with the
"custom:<id>" entries (legacy single-custom mode unchanged).
Co-authored-by: Cursor <cursoragent@cursor.com>
Two independent robustness fixes:
1. ChatChannel.cancel_session / cancel_all_session raised KeyError when a
session existed in self.sessions but no future had been dispatched yet.
self.sessions[sid] is created in produce(), but self.futures[sid] is only
created later in consume() on first dispatch. Cancelling in that window
(e.g. user sends a message then immediately cancels) crashed the cancel
path. Use self.futures.get(sid, []) so an absent entry is a no-op.
2. compress_imgfile decremented JPEG quality by 5 with no lower bound. For an
image that cannot be compressed below max_size, quality went 0, negative,
... — the loop never terminated and passed invalid quality values to PIL.
Add a min_quality floor (10) and return the best effort once reached.
Adds tests/test_robustness_fixes.py covering both paths (5 tests).
