English · 简体中文

ADS Agent Bridge

From an exact ADS selection or blank circuit to checked data and an editable DDS result.

Finish a circuit-to-results task in one conversation

“Start from a blank workspace, build this testbench, run the simulation, export the data, and leave the response plot editable in DDS.”

Built in ADS	Result left editable in DDS

The public ADS 2026 Update 2.1 acceptance completed the whole request:

• built a six-instance AC circuit from an empty workspace;
• ran the circuit simulation and returned 31 finite rows;
• exported CSV and created two native DDS pages with rectangular and polar plots;
• saved, closed, and freshly reopened the editable result;
• completed the four-stage Runtime plan in 4.312 seconds.

These are real ADS application-window captures; the DDS curve remains native and editable. The same maintained path can continue from an exact schematic, layout, library-tree, or DDS selection copied with Copy ADS Context, and can run an already-generated Momentum input on a protected sibling copy.

ADS Agent Bridge connects Codex or Pi Agent to the ADS installation and object you actually selected. Version-matched local documentation, the DE/DDS context plug-in, controlled live sessions, and bounded automation stay with ADS on the EDA host; repeated remote work reuses EDA Bridge Runtime instead of rebuilding an SSH command for every action.

New ADS functionality does not require a new Bridge wrapper. The Agent first uses version-matched official docs and a small packaged experience library, then runs official ADS Python through a governed workspace transaction. Maintained operations such as design.apply and dds.create remain available as asset-bound compiled shortcuts: they save tokens and transcription errors, but never define the outer limit of ADS capability.

Successful governed native work returns an opaque continuation Context. A later batch can reuse its exact host-private target and content fingerprint; the Bridge still requires a new explicit program, effect, write scope, purpose, idempotency key, and validation, and rejects stale content or conflicting identity. See the continuation contract.

Start in three steps

Prerequisites: a licensed ADS installation, Windows or Linux, and Python 3.10 or later for the public command.

pipx install ads-agent-bridge
ads-agent setup
ads-agent quickstart

Installing the package automatically installs its compatible eda-bridge-runtime Python dependency. You do not need to install a second Python package by hand. If the Agent runs on another computer, enable the EDA Bridge Runtime MCP/plugin there; the ADS-only host does not need the Agent-facing plugin.

setup discovers installed ADS versions, asks for an explicit selection, and installs the recoverable context add-on and two cooperating Skills. quickstart passes only after documentation query, add-on registration, disposable workspace creation, minimal circuit simulation, and dataset readback have each passed.

If pipx or a suitable Python is not yet available, use the versioned bootstrap for Linux or Windows PowerShell. The bootstrap creates an isolated environment and does not replace an externally managed system Python.

Then open an exact workspace:

ads-agent --pretty launch --workspace /path/to/MyWorkspace_wrk
ads-agent --pretty status
ads-agent disconnect

disconnect leaves ADS running. shutdown requests native exit only for a matching Agent-owned session.

What you can ask your Agent

The capability matrix gives the maintained evidence and stop rule behind every row.

The selection plug-in

After ADS restarts, Copy ADS Context is available from supported DE schematic, layout, symbol, folder/library-tree, and DDS selections. The copied text identifies the software, host-local origin, workspace, object kind, selection, and freshness needed by the Agent. It contains no password, live port, or permission to mutate.

This is the normal user interaction:

1. Select the intended object in ADS.
2. Click Copy ADS Context.
3. Paste it into the conversation and describe the task naturally.
4. Review the target and evidence returned by the Agent.

See the context interaction contract for exact selection coverage.

Public evidence

The maintained acceptance path uses real ADS installations on Windows and Linux. It separately checks documentation, context capture, live session identity, safe dialog supervision, typed schematic construction, circuit simulation, dataset and CSV readback, native DDS equation and plot creation, and generated-input Momentum execution.

The maintained blank-workspace → schematic → simulation → native DDS path passed as one four-stage Runtime plan in 4.312 seconds, with 31 finite rows, a deterministic native dataset, CSV, and a freshly reopened two-page DDS report containing rectangular and polar plots. See the sanitized workflow evidence.

Two narrow ADS 2027 comparisons are also public:

• In nine isolated knowledge tasks, Bridge completed 9/9, used 10.4% fewer total tokens, and had 5.2% higher median latency than the official ADS MCP. Method and data
• In a three-run headless AC microbenchmark, both paths completed 3/3; Bridge used 43.4% fewer total tokens and had 21.5% lower median wall time. Method and data

These are small, isolated regression suites, not a universal product ranking. They do not compare installation, the DE/DDS plug-in, GUI session control, dialog handling, or every ADS solver workflow; this is not a full-product comparison. The sanitized headless per-run summary is also available as JSON.

Local and remote use follow one path

On a remote ADS host, repeated Agent operations use:

ads-agent runtime serve

EDA Bridge Runtime keeps one SSH stdio process alive, records the purpose and timing of each operation, and never exposes the embedded ADS Bridge port. If Agent and ADS share a machine, register the same service as a local connection; do not bypass Runtime. The engineering behavior and evidence remain the same in both topologies.

Safety and privacy

• Local documentation, indexes, workspaces, tokens, and results stay on the EDA host unless the user deliberately exports them.
• Live endpoints bind to loopback and require a random session token.
• A reused session must match the selected ADS installation and exact workspace.
• A Context identifies the target but never authorizes editing or simulation.
• Structured edits protect the source and require fresh-reopen assertions.
• Dialog actions require a fresh process/window fingerprint, not screen coordinates.
• Shutdown refuses unverified or user-owned sessions and never force-kills ADS.
• Arbitrary embedded Python and dynamic AEL remain disabled by default.

Support and boundaries

Next

• governed access to version-matched official Python/AEL so new documented ADS uses do not require one Bridge wrapper per component, plot, or solver option;
• richer RF testbenches, reusable parameterized cells, native DDS, layout, and EM work delivered through that common path and selectively promoted as certified workflows.

More information

• Installation and command reference
• Five public examples
• Capability and evidence matrix
• Architecture and capability growth
• Operation classification
• Session and dialog behavior
• Execution context contract
• Release contract
• Changelog

To remove only this ADS integration while preserving unrelated add-ons:

ads-agent addon uninstall