chore: upgrade fork to latest upstream Multica

.env.example

@@ -167,6 +167,20 @@ MULTICA_LLM_DEFAULT_MODEL=
 #     other 4xx. This budget does not cover a stream that breaks mid-response,
 #     and it is unrelated to task-level retries.
 MULTICA_LLM_MAX_RETRIES=
+#   - Ask an OpenAI-compatible gateway to turn the model's reasoning
+#     ("thinking") off by adding chat_template_kwargs: {"enable_thinking":
+#     false} to every request body this layer sends. Some gateways and
+#     GLM/Qwen-style model routes honor the field; there it removes the
+#     thinking pass that would otherwise dominate the latency budget of the
+#     assist calls above. Standard OpenAI endpoints reject unknown body
+#     fields, so only enable this when the upstream accepts it. GPT-5.6-family
+#     models already get reasoning_effort=none on the follow-up questions
+#     request. Chat auto-titling sends no reasoning field; this switch affects
+#     it only when the configured upstream accepts chat_template_kwargs.
+#     Accepted values are true/false and 1/0
+#     (case-insensitive); anything else fails the boot. The startup log
+#     ("llm retry policy") reports the effective state.
+MULTICA_LLM_DISABLE_THINKING=
 MULTICA_DAEMON_CONFIG=
 MULTICA_WORKSPACE_ID=
 MULTICA_DAEMON_ID=
@@ -623,6 +637,19 @@ MULTICA_LARK_WS_PROXY_URL=
 # Generate one with: openssl rand -base64 32
 MULTICA_DINGTALK_SECRET_KEY=
 
+# Slack bot integration (Settings → Integrations → Slack)
+# Off until MULTICA_SLACK_SECRET_KEY is set — a base64-encoded 32-byte key
+# that encrypts each installation's bot and app tokens at rest. Leave empty
+# to disable. Generate one with: openssl rand -base64 32
+MULTICA_SLACK_SECRET_KEY=
+
+# Telegram bot integration (Settings → Integrations → Telegram)
+# Off until MULTICA_TELEGRAM_SECRET_KEY is set — a base64-encoded 32-byte key
+# that encrypts each Bot's token at rest. Leave empty to disable. Keep it
+# stable: losing or rotating it makes stored Bot tokens unreadable.
+# Generate one with: openssl rand -base64 32
+MULTICA_TELEGRAM_SECRET_KEY=
+
 # Frontend
 # Leave empty — auto-derived from page origin in browser, set by Makefile for local dev.
 # NEXT_PUBLIC_API_URL also feeds the Next.js SSR proxy when explicitly set.
@@ -640,16 +667,18 @@ NEXT_PUBLIC_WS_URL=
 # REMOTE_API_URL=https://multica-api.copilothub.ai
 
 # ==================== Self-hosting: Control Signups (fixes #930) ====================
-# Set to "false" to completely disable new user signups (recommended for private instances)
+# Set to "false" to restrict new accounts to allowlisted or invited users.
+# A pending, unexpired workspace invitation also permits emails outside either
+# allowlist when ALLOW_SIGNUP=true. Revocation does not remove existing accounts.
 ALLOW_SIGNUP=true
 # The web UI reads ALLOW_SIGNUP from /api/config at runtime, so toggling this
 # only requires restarting the backend / compose stack — not rebuilding web.
 # It is not hot-reloaded.
 
-# Optional: Only allow emails from these domains (comma-separated)
+# Optional: Allow emails from these domains without an invitation (comma-separated)
 ALLOWED_EMAIL_DOMAINS=
 
-# Optional: Only allow these exact email addresses (comma-separated)
+# Optional: Allow these exact email addresses without an invitation (comma-separated)
 ALLOWED_EMAILS=
 
 # Set to "true" to disable workspace creation for every caller on this

.github/ci-paths.json

@@ -102,6 +102,7 @@
     "Makefile",
     "server/internal/handler/reserved_slugs.json",
     "packages/core/paths/reserved-slugs.ts",
+    "server/cmd/server/router.go",
     ".github/workflows/ci.yml",
     ".github/ci-paths.json",
     "scripts/ci-scope.mjs",

.github/workflows/ci.yml

@@ -282,6 +282,14 @@ jobs:
   # release, an entry built on the old one must not be restored. setup-go
   # keys on ImageOS for the same reason (actions/setup-go#368).
   #
+  # A rolling entry only grows on its own: each save is the restored entry
+  # plus this commit's objects, and Go trims only what has sat unused for
+  # five days, so the entry carried every recent main commit's race objects.
+  # It reached 3.5 GB in ten days, and restoring and saving it cost more than
+  # the compile it saved (MUL-7547). Before saving, the job therefore drops
+  # every entry this run did not use, leaving the next run the objects of the
+  # tree it is most likely to build.
+  #
   # This job is the only writer of either entry. backend-agent-tests restores
   # both, but it downloads just pkg/agent's slice of the module graph and
   # finishes first, so letting it save would repeat the first-writer-wins
@@ -365,6 +373,14 @@ jobs:
           key: ${{ steps.gocache.outputs.prefix }}${{ github.sha }}
           restore-keys: ${{ steps.gocache.outputs.prefix }}
 
+      # Go refreshes a cache entry's mtime when it uses it, but only once that
+      # mtime is an hour old. So after the run, an entry no newer than an hour
+      # before the first compile is one this run did not use. The prune step
+      # before the save compares against this marker.
+      - name: Mark Go build cache epoch
+        id: goepoch
+        run: touch -d '-1 hour' "$RUNNER_TEMP/go-build-epoch"
+
       - name: Build
         run: cd server && go build -race ./...
 
@@ -387,6 +403,17 @@ jobs:
         # pkg/agent runs on its own runner: backend-agent-tests below.
         run: bash scripts/test-go.sh --race --only regular
 
+      # Go's own trim (cmd/go/internal/cache) with this run as the window
+      # instead of five days: only `-a`/`-d` entries, and an executable entry
+      # is a directory, so it goes whole. See the job comment.
+      - name: Prune Go build cache to this run's entries
+        if: ${{ !cancelled() && github.event_name == 'push' && steps.goepoch.outcome == 'success' }}
+        run: |
+          du -sh ~/.cache/go-build
+          find ~/.cache/go-build -mindepth 2 -maxdepth 2 \( -name '*-a' -o -name '*-d' \) \
+            ! -newer "$RUNNER_TEMP/go-build-epoch" -exec rm -rf {} +
+          du -sh ~/.cache/go-build
+
       # Publishes from main only; see the job comment. `!cancelled()` so a red
       # main still refreshes the objects for the next run, and the key step's
       # outcome so a run that never reached Go does not save an empty entry
@@ -609,6 +636,45 @@ jobs:
         # silently as "ok".
         run: go test ./internal/daemon/execenv -v -run '^TestWindowsOpenclawShim' -count=1 -timeout=5m
 
+      - name: Test Windows workdir containment guard in the CLI
+        if: ${{ needs.changes.outputs.backend == 'true' }}
+        working-directory: server
+        # The CLI's --content-file / --attachment guard decides whether a path is
+        # inside the task workdir by canonicalizing both sides. Two of the things
+        # that decision rests on are Windows-only and covered nowhere else: a
+        # candidate on another volume, where filepath.Rel returns an error rather
+        # than a "..", so the guard has to read that as "outside" instead of
+        # surfacing `Rel: can't make Z:\... relative to C:\...` as a resolve
+        # failure; and a directory junction — the containment-relevant link shape
+        # that needs no elevation, so the one real hosts have — which resolves to
+        # itself rather than to its target under this module's go directive. The
+        # junction test pins that gap deliberately, records the observations it
+        # rests on, and fails with instructions if the platform ever starts
+        # resolving them.
+        # cmd/multica has no Windows job otherwise, so these windows-tagged
+        # tests run nowhere else. It matters twice over on a non-admin host,
+        # where the package's cross-platform cases skip for want of the symlink
+        # privilege; the ubuntu backend job covers those shapes instead.
+        # -v so a silent skip (or a -run pattern that stops matching) is visible
+        # in the log instead of passing as "ok".
+        run: go test ./cmd/multica -v -run '^TestFileWithinWorkingDirWindows' -count=1 -timeout=5m
+
+      - name: Test Windows link-target classification in internal/util
+        if: ${{ needs.changes.outputs.backend == 'true' }}
+        working-directory: server
+        # The guard's Windows correctness rests on internal/util's link-target
+        # grammar (classifyTarget), whose branch ordering is load-bearing:
+        # VolumeName is non-empty for drive-absolute paths too, so asking it
+        # before IsAbs misreads a junction's absolute target as drive-relative
+        # and resolves it against the working directory — exactly how a
+        # junction escape read as inside the workdir for one CI run. The
+        # fixture-free table pins every branch without needing a filesystem,
+        # but windows-tagged util tests run nowhere else. Scoped to the table
+        # by name; the package's symlinked-workdir cases belong to the ubuntu
+        # backend job, which runs the whole package.
+        # -v so a silent skip or a pattern that stops matching is visible.
+        run: go test ./internal/util -v -run '^TestClassifyTarget' -count=1 -timeout=5m
+
       - name: Test Windows isolated repo checkout is committable
         working-directory: server
         # #6449: on Windows the daemon now hands Codex tasks a checkout whose
@@ -670,7 +736,7 @@ jobs:
         # explicitly configured path reach the release executable. The same job
         # covers the npm shape, where the entry point is a `.cmd` shim that only
         # the command interpreter can run.
-        run: go test ./internal/daemon -v -run '^(TestCanonicalExecutablePath|TestTrimExtendedLengthPrefix|TestResolveAgentExecutablePathKeeps|TestResolveAgentEntry(FollowsRetargetedInstallerJunction|CanonicalizesRediscoveredJunction|ForLaunchKeepsCmdShimLaunchable|ForLaunchRejectsUnverifiedInitialJunctionTarget|ForLaunchRejectsRediscoveredJunctionWhenFinalPathResolutionFails|DoesNotSharePreRetargetSingleflightResult|ForLaunchFailsWhenJunctionKeepsRetargeting)|TestHandleTaskReportsWindowsCodexProcessStartFailure)' -count=1 -timeout=5m
+        run: go test ./internal/daemon -v -run '^(TestCanonicalExecutablePath|TestTrimExtendedLengthPrefix|TestResolveAgentExecutablePathKeeps|TestResolveAgentExecutablePath_ProfileOverride|TestResolveAgentEntry(FollowsRetargetedInstallerJunction|CanonicalizesRediscoveredJunction|ForLaunchKeepsCmdShimLaunchable|ForLaunchRejectsUnverifiedInitialJunctionTarget|ForLaunchRejectsRediscoveredJunctionWhenFinalPathResolutionFails|DoesNotSharePreRetargetSingleflightResult|ForLaunchFailsWhenJunctionKeepsRetargeting)|TestHandleTaskReportsWindowsCodexProcessStartFailure)' -count=1 -timeout=5m
 
       - name: Build Windows CLI helper entrypoint
         working-directory: server
@@ -777,7 +843,12 @@ jobs:
         if: runner.os != 'Windows'
         run: bash scripts/install.test.sh
 
-      - name: Test PowerShell installer
+      - name: Test installer with Windows PowerShell 5.1
+        if: runner.os == 'Windows'
+        shell: powershell
+        run: ./scripts/install.ps1.test.ps1
+
+      - name: Test installer with PowerShell 7
         if: runner.os == 'Windows'
         shell: pwsh
         run: ./scripts/install.ps1.test.ps1

.github/workflows/coderpush-images.yml

@@ -0,0 +1,52 @@
+name: CoderPush main images
+
+# Build immutable artifacts from main. Publishing images does not deploy them.
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: coderpush-images-${{ github.sha }}
+  cancel-in-progress: false
+
+jobs:
+  build:
+    if: github.repository == 'CoderPush/multica' && github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - image: multica-backend
+            dockerfile: Dockerfile
+          - image: multica-web
+            dockerfile: Dockerfile.web
+    steps:
+      - uses: actions/checkout@v4
+      - uses: docker/setup-buildx-action@v3
+      - uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      - uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ${{ matrix.dockerfile }}
+          platforms: linux/amd64
+          push: true
+          tags: ghcr.io/coderpush/${{ matrix.image }}:sha-${{ github.sha }}
+          labels: |
+            org.opencontainers.image.source=https://github.com/CoderPush/multica
+            org.opencontainers.image.revision=${{ github.sha }}
+          build-args: |
+            VERSION=main-${{ github.sha }}
+            COMMIT=${{ github.sha }}
+            NEXT_PUBLIC_APP_VERSION=main-${{ github.sha }}
+          cache-from: type=gha,scope=${{ matrix.image }}
+          cache-to: type=gha,mode=max,scope=${{ matrix.image }}

.gitignore

@@ -79,3 +79,5 @@ data/
 __pycache__/
 *.pyc
 perf-report/
+# The production example is a tracked template; `.env*` above would hide it.
+!apps/mobile/.env.production.example

CLI_AND_DAEMON.md

@@ -31,7 +31,23 @@ For install script or manual installs, use:
 multica update
 ```
 
-`multica update` auto-detects your installation method and upgrades accordingly.
+`multica update` uses GitHub Releases by default. Self-hosted installations can
+point the CLI at a GitHub Releases-compatible metadata mirror and an artifact
+mirror without changing the command. To let a daemon poll that source, also
+enable self-update explicitly because self-hosted auto-update is disabled by
+default:
+
+```bash
+export MULTICA_RELEASE_API_BASE_URL=https://updates.example/api
+export MULTICA_RELEASE_DOWNLOAD_BASE_URL=https://updates.example/releases/download
+export MULTICA_DAEMON_AUTO_UPDATE=true
+multica update
+```
+
+The metadata mirror must serve `/repos/multica-ai/multica/releases/latest` and
+`/repos/multica-ai/multica/releases/tags/<tag>`. The artifact mirror must serve
+`/<tag>/<asset-name>` and preserve the published `checksums.txt` contents.
+When these variables are unset, the GitHub defaults remain unchanged.
 
 ## Quick Start
 
@@ -1101,3 +1117,11 @@ On the API, both endpoints accept `?include=content` and `?include=metadata`.
 A request that sends neither still gets `content`, on both endpoints, so a
 server upgrade never changes what an un-upgraded client receives — it is the
 CLI that asks for the smaller shape.
+
+### Custom runtime compatibility targets
+
+Create custom Oh-My-Pi profiles with `multica runtime profile create --runtime-type omp --command-name omp --display-name "Custom Oh-My-Pi"`.
+The immutable `runtime_type` selects model discovery, skills paths, and launch behavior;
+the server derives `protocol_family` (`pi` for `omp`). Custom command/path overrides and
+fixed arguments still apply, and the runtime retains its custom-profile provenance.
+Existing profiles and the legacy `--protocol-family` flag retain their original target.

LICENSE

@@ -12,6 +12,11 @@ together and what must be delivered when you redistribute.
 
                        Part I — Additional Conditions
 
+In this Multica License, "the producer" means Index Labs (Hong Kong)
+Limited, the company that develops and distributes Multica. Commercial
+licenses and branding waivers may be requested at
+https://www.multica.ai/contact-sales.
+
 1. Multica may be utilized commercially, including as a backend service for
    other applications or as a task management platform for enterprises,
    subject to the following conditions:
@@ -112,7 +117,7 @@ together and what must be delivered when you redistribute.
       deliver this complete file. Delivering Part II alone does not
       satisfy the Multica License or section 4(a) of Part II.
 
-© 2025-2026 Multica, Inc.
+© 2025-2026 Index Labs (Hong Kong) Limited
 
 ------------------------------------------------------------------------
 

NOTICE

@@ -1,8 +1,8 @@
 Multica
-Copyright 2025-2026 Multica, Inc.
+Copyright 2025-2026 Index Labs (Hong Kong) Limited
 
-This product includes software developed at Multica, Inc.
-(https://github.com/multica-ai/multica).
+This product includes software developed at Index Labs (Hong Kong) Limited
+for Multica (https://github.com/multica-ai/multica).
 
 Multica is distributed under the Multica License, which incorporates the
 complete text of the Apache License, Version 2.0 together with additional

README.md

@@ -10,7 +10,7 @@
 
 **Agents that show up on the board.**
 
-Multica is an open-source workspace where you assign work to AI coding agents the way you'd
+Multica is a source-available workspace where you assign work to AI coding agents the way you'd
 assign it to a teammate — they pick up the issue, report progress, raise blockers, and hand it
 back for review. Self-hostable, works with 26 agent CLIs, no lock-in.
 

README.zh.md

@@ -10,7 +10,7 @@
 
 **智能体，也在看板上。**
 
-Multica 是一个开源的团队工作区。你像给同事派活一样，把任务交给 AI 编码智能体——它自己接手、边做边
+Multica 是一个源码公开的团队工作区。你像给同事派活一样，把任务交给 AI 编码智能体——它自己接手、边做边
 汇报、卡住了主动说，做完交回来给你审。可自部署，支持 26 种智能体 CLI，不绑定任何厂商。
 
 [![CI](https://github.com/multica-ai/multica/actions/workflows/ci.yml/badge.svg)](https://github.com/multica-ai/multica/actions/workflows/ci.yml)
@@ -245,7 +245,7 @@ iOS 客户端在 [`apps/mobile/`](apps/mobile/)，怎么编译装到自己 iPhon
 
 ---
 
-## 开源协议
+## 许可协议
 
 [Multica License](LICENSE) —— Apache License 2.0 全文并入，外加针对托管服务、商业嵌入和品牌标识的
 附加条件。自部署、改代码、在它之上做东西都可以；准确条款以 [LICENSE](LICENSE) 为准，署名信息见

SELF_HOSTING.md

@@ -263,6 +263,8 @@ When developing from a checkout, use the local chart path instead:
 helm install multica deploy/helm/multica -n multica
 ```
 
+If the backend has to reach services whose certificates come from an internal CA, such as a self-hosted Gitea, set `backend.extraCACerts.configMap`. See [Advanced Configuration → Private CA Certificates](SELF_HOSTING_ADVANCED.md#private-ca-certificates-optional).
+
 Watch the pods come up:
 
 ```bash