AI-assisted AUTOSAR review and architecture history.

01 / The Problem

You already know the meeting.

03 / What we're building

Treat ARXML like the source code it is.

• 01

  Review architectural changes asynchronously, not in meetings.

  When someone updates an SWC, port, or composition, reviewers see what actually changed at a glance. A visual, structured summary of every architectural impact, instead of thousands of lines of XML. Works inside the pull request review you already use.

  Technically: Semantic ARXML diffs at the element level, with optional inline architectural visualization, posted as PR comments via your CI of choice.

• 02

  Stop saying "nobody knows when this changed."

  When did this port get added? Who changed the timing on this runnable? Why does this interface look the way it does? Ask your repository in plain English and get the answer in seconds. AI-assisted summarization compresses years of commits into a readable narrative when you need context. The same history layer produces clean change reports between any two ECU releases for ISO 26262 and A-SPICE audits.

  Technically: Element-level history and blame, indexed across the entire Git history. Natural-language queries and summaries run on your own BYOK LLM endpoint. Your ARXML never leaves your infrastructure.

• 03

  Your architecture stays yours.

  No uploads. No cloud processing. No third party ever sees your ARXML. The product runs locally on each engineer's machine and on your own CI runners. Self-hosted deployment available for teams that need it.

  Technically: Local-first CLI, desktop app, and VS Code extension. AI features (optional) connect to your own Azure OpenAI or AWS Bedrock. Never to ours.

04 / For the engineers who scroll this far

When did this change? Who made it? Finally, an answer.

Standard git blame on an ARXML file is line-by-line and meaningless. Line numbers shift every time the upstream editor regenerates the XML. The Manifold equivalent shows you every architectural change to the element you asked about, with author, date, and the actual semantic delta.

$ git diff HEAD~1 PowertrainComp.arxml | head -32

diff --git a/Powertrain/PowertrainComp.arxml b/Powertrain/PowertrainComp.arxml
index a3f2d8c..a8f2e91 100644
--- a/Powertrain/PowertrainComp.arxml
+++ b/Powertrain/PowertrainComp.arxml
@@ -14820,15 +14820,8 @@
                       <DATA-ELEMENT-REF DEST="VARIABLE-DATA-PROTOTYPE">
                         /AUTOSAR/DataTypes/Temperature_T
                       </DATA-ELEMENT-REF>
-                      <INIT-VALUE>
-                        <CONSTANT-REFERENCE>
-                          /AUTOSAR/Constants/InitTemp
-                        </CONSTANT-REFERENCE>
-                      </INIT-VALUE>
                     </NONQUEUED-RECEIVER-COM-SPEC>
@@ -15239,12 +15232,12 @@
                 <TIMING-EVENTS>
                   <TIMING-EVENT>
                     <SHORT-NAME>UpdateBatteryState_Event</SHORT-NAME>
-                    <PERIOD>0.1</PERIOD>
+                    <PERIOD>0.05</PERIOD>
                   </TIMING-EVENT>
@@ -28104,9 +28097,3 @@
                   </P-PORT-PROTOTYPE>
-                  <R-PORT-PROTOTYPE UUID="d2ae..7c91">
-                    <SHORT-NAME>InPort_TempSensor</SHORT-NAME>
-                    <REQUIRED-INTERFACE-TREF DEST="SR-INTERFACE">
-                      /AUTOSAR/Interfaces/TempReading_IF
-                    </REQUIRED-INTERFACE-TREF>
-                  </R-PORT-PROTOTYPE>
                 </PORTS>

                                ... 4,213 more lines changed

$ manifold blame /Powertrain/MotorControl/BatteryManager
 
History of SWC: BatteryManager  (5 changes in last 18 months)
 
────────────────────────────────────────────────────────
2025-02-14  Tomas L.    cleanup: remove InPort_TempSensor
            a8f2e91     − R-PORT InPort_TempSensor
                          interface: TempReading_IF (SR)
 
2024-09-12  Lukas K.    feat: tighten timing on update_battery_state
            7d4e9a1     ~ runnable.update_battery_state
                        period: 100ms → 50ms
 
2024-06-03  Anna M.     feat: expose battery status to vehicle bus
            b7e4f12     + P-PORT OutPort_BatteryStatus
                          interface: BatteryStatus_IF (SR)
                        + R-PORT InPort_TempSensor
                          interface: TempReading_IF (SR)
 
2024-01-18  Tomas L.    refactor: extract from PowertrainComp
            c2a1b9d     extracted from BatteryControl composition
 
2023-11-04  Lukas K.    initial: BatteryManager SWC
            a3f2d8c     + 2 ports, 1 runnable, BSW dep: NvM
────────────────────────────────────────────────────────
 
$ 

Same Git repository. Same commits. Same ECU target. The difference is whether your tools understand what a port, a runnable, and a composition actually mean.

05 / Plays nice with

For teams already using:

• SystemDesk ·
• DaVinci Developer ·
• DaVinci Configurator ·
• ISOLAR-A / ISOLAR-AB ·
• EB tresos

Manifold operates on the ARXML your existing toolchain already produces and the generated RTE / BSW code that comes out of it. It does not replace SystemDesk, DaVinci, ISOLAR, or EB tresos. Your architects keep using the editor they know; Git, finally, becomes useful.

07 / Common questions

The questions we get asked first.

• Does my ARXML get uploaded anywhere?

  No. Manifold runs locally: CLI, desktop app, VS Code extension, and on your own CI runners. We never see file contents. Self-hosted deployment is available for teams that require it by policy.
• Do your AI features see my data?

  No. AI features connect to your own LLM endpoint: Azure OpenAI, AWS Bedrock, Vertex, or a self-hosted model. We never see prompts, responses, or ARXML. AI can be disabled entirely and the rest of the product still delivers its core value.
• How is this different from dSPACE AUTOSARCompare?

  AUTOSARCompare is a desktop tool for on-demand comparison between two file versions, run by a single engineer when they need it. Manifold is Git-native. It lives inside the pull request review you already use, indexes element-level history across the entire repo, and produces structured PR comments for async team review. Different shape of tool for different jobs; many teams will end up using both.
• Which Git hosts does it support?

  Any Git host. The semantic engine and CLI are platform-agnostic. The CI integrations ship as a GitHub Action, GitLab CI job, Bitbucket Pipeline, and Azure DevOps task. Each posts the same semantic diff summary as PR/MR comments. If your team self-hosts Gitea or runs an on-prem GitLab, the CLI still works.
• What about Adaptive AUTOSAR?

  Yes. The semantic engine handles ARXML for both Classic and Adaptive. AI-assisted Classic→Adaptive migration is one of the called-out capabilities for teams moving toward SDV and zonal architectures.
• When can I actually try it?

  Early access is opening soon. Sign up above and we will send a single email when it is ready. No marketing in between. If you would like to be a design partner or beta tester sooner, reply to the confirmation email and tell us about your team; we are talking to a small number of teams during build.