Changelog
Everything we have shipped, most recent first. The landing page shows the latest five; this is all of it.
Three new calculators: LDO thermal, boost converter, power budget
Three more free calculators, no account needed. The LDO calculator gives the dissipation at each input voltage, the junction temperature and the heat sink the part needs. The boost converter calculator sizes the inductor and output capacitor and checks the peak switch current against the chip's current limit. The power budget adds up a tree of regulators into source current, losses and battery runtime; each LDO in it opens the LDO calculator with its numbers filled in.
Try Pro or Team free for a week
The pricing plans and the ROI article now offer "Start a free trial". Pick Pro, for one engineer, or Team, for a group sharing designs, then sign in, choose monthly or yearly, and go through checkout. The trial lasts 7 days, one per person: checkout takes a card but charges nothing until the trial ends, and cancelling before then costs nothing. If you have already used your trial, the plan now says so up front instead of promising one.
Connect your AI assistant on a free trial
When you connect Claude or another assistant to Electronics Architect from a free account, the sign-in page used to stop at "upgrade, then start again". It now offers Pro's 7-day free trial right there: pick monthly or yearly, go through checkout, and you land back on the same page to finish connecting, with no need to start over in the assistant. Nothing is charged during the trial, and cancelling before it ends costs nothing. Opening checkout and backing out no longer uses up your trial, either.
Every report PDF now reads the same way
The derating report and the combined dossier have been rebuilt to match the tolerance report: the same summary that states the finding up front, the same table ranking every part by its worst measurement, the same plain-English glossary, and the same three verdicts — over the limit, close to it, or clear. The derating report gains a chart it never had, one bar per measurement running from zero to the part's own rating with the limit marked on it, so you can see how much margin is left instead of reading a percentage. One correction worth knowing if you printed a tolerance report yesterday: bars that straddled a spec limit were coloured as failures even when no trial had actually failed, so a clean run could look like it had problems. Bars now split exactly on the limit, and red means a real trial went outside it.
The tolerance report now tells you what it found
The PDF you get after a tolerance run used to be a page of numbers per part and not much else. It now opens with the finding written out in a sentence or two — what held, what is tight, and what to change first — above a table that ranks every part by its worst measurement. Each measurement gets one of three verdicts instead of a bare number: it went outside its limit, it stayed inside but with no room to spare, or it is clear. The second one matters, because a part that passed this run with nothing in hand will not keep passing. Every chart is labelled now, both axes, with the limits drawn on it and a short glossary on the second page explaining each term the report uses.
Long tolerance runs from your AI assistant
The long runs added earlier today — the ones that check how a design behaves when real parts vary from their spec sheets, and that take longer than a single request allows — can now be asked for from an assistant connected to our MCP server, not only from a script. It starts the run, checks on it every ten seconds, and after the first check tells you how many more it expects, measured from the pace so far, so you hear "about two minutes" or "about an hour" before you commit to waiting. The result is the same one the one-shot run gives.
Monte Carlo runs that are too big for one request
A tolerance analysis on a large design can take longer than any single web request is allowed to live, so those runs simply could not be asked for. Now they can: POST /v1/jobs/monte-carlo starts a run, does the first ten seconds of trials straight away, and hands back a job id with its progress. Each time you ask after it, another ten seconds of trials get done and the reply tells you how far along it is — so the run advances as you follow it, and nothing is lost if it pauses in between. Ask for a small run and it just finishes in the first reply, exactly as before. A job at a given seed returns the same statistics as the one-shot endpoint at that seed, to the last decimal. Two runs at a time per account, results kept for a day.
Derating through the API and your AI agent
The derating analysis the editor runs — every rated part held to NASA GSFC or IPC-9592 after a nominal solve — is now a call: POST /v1/designs/derating for scripts, run_derating for an assistant on the MCP server. Name the standard and get back, per part, each stress parameter's actual, rated and derated limit, a pass / marginal / over verdict and what to change. It is the same evaluator the Derating panel uses, so the agent's report and the one on your screen cannot disagree. Unmetered, like solving.
The page outline is the sheet your PDF prints on
Turn it on in Canvas → Page setup and the canvas dims everything outside the A4, Letter or 16:9 sheet that Export Studio will produce, with the frame, the zone references and the title block drawn where the PDF puts them, so the outline on screen is the page you get. It is off by default now, in every template too. The toolbar tidied up along the way: File holds the bill of materials and Share, Canvas holds Page setup, and Export is one button that opens Export Studio. The component palette works without a mouse — Tab to a tile and press Enter to place it in the middle of the view, selected. And the Help Center's screenshots show the editor as it is today.
Everything we have shipped, on one page
This list used to scroll inside a box on the landing page, and the box caught your mouse wheel on the way down. Now the front page shows the latest five changes and the changelog keeps the whole history on a page of its own.
A developers page, and the site in one file for agents
/developers now shows the three ways to connect an assistant — the MCP address for Claude Desktop and Claude Code, a key pasted as a bearer header for Antigravity and Cursor, and the community mcp-remote bridge for clients that only run local servers — with the config block for each, the REST path, and the three workflows our acceptance runs actually performed. Every tool the server offers now tells the client whether it only reads or can replace something, so a well-behaved assistant asks you before an overwrite and not before a lookup. For agents that read documentation directly, /llms-full.txt carries the developer pages, the articles and the calculators in one file, generated from the pages so it cannot fall behind them. The story of how this came to be is in Let your agent do the power tree.
Hand a design to your AI agent, and get it back on the canvas
The hand-off now works in both directions. When an agent creates or changes a design through the API or the MCP server, the link it gives you opens that design straight on the canvas — sign in if you need to — with a quiet banner saying which API key made it and when, and See what changed shows the previous state with Restore one click away. The other way round, Share → Copy MCP snippet puts a short instruction on your clipboard that any agent connected to your account can use to fetch the design by id; Copy as JSON gives the document itself. And every warning in the panel now shows the same code the API returns, so "clear the two thermal warnings on L1 and L2" means the same thing to you and to your agent.
Top-up packs, and 250 Monte Carlo sims a month on Pro
Pro's Monte Carlo allowance goes from 100 to 250 sims a month. And when a month's allowance does run out, Pro and Team accounts no longer wait for the reset: the limit dialog offers a $5 top-up pack — 50 Monte Carlo sims, 20 AI analyses, or 5 AI Synthesis generations with 50 refinements — paid once through Stripe, added to the account, never expiring, and used only after the monthly allowance. Nothing recurs and the plan does not change. An AI assistant working through the API sees the pack balance on get_account_usage and, at the limit, is told what would unblock it.
See what each API key is doing, and cap it
Account → Developer now shows, for every key, how many calls it made today and over the last 30 days, split by what they were — solving, exporting, AI review. Set a key's requests-per-minute limit when you create it and change it later without minting a new one. On a Team plan the admin issues keys for the team from the same tab and sees the same figures for each of them. An agent can read its own key's counts from GET /v1/me.
Every export says what solved it, and can be checked
Every PDF, PNG, BOM and JSON export now ends with one line: the solver release that produced it, a fingerprint of the design as solved, the date, and a verify link. Anyone handed the report can open the link, paste the design (or follow a public share link) and see the same solver re-solve it and confirm the fingerprint. For agents and scripts the same release adds compare_designs — what changed between two designs and what it did to loss, efficiency, warnings and yield — plus solve_json and report_json exports, and every warning that has a one-property answer now carries a fix patch with the number in it ("choose a part rated ≥ 2.9 A").
One AI design review instead of two buttons
Explain and Suggest asked the same reviewer the same question twice and charged you for both. They are now one button, AIDR (AI Design Review): in a single pass it says what the solved numbers mean — the efficiency losses, thermal margin, dropout headroom and reliability risks, by designator — and then what to change, with the estimated gain. One review, one AI call. At the bottom of every review is a link that posts the design to Community Design Review, for the opinion only another engineer can give. The API and the MCP server follow: POST /v1/ai/review and the review_design tool replace the old pair.
Agents can solve a saved design by id and ask for a compact result
An AI assistant driving the tool no longer re-sends a whole design for every tolerance corner: solve_design and the Monte Carlo tool take the id of a design saved in your account, and compact: true returns the solved numbers and one grouped warning list at a quarter to a third of the size. A get_template tool reads any shipped template as a document. Negative rails — a −5 V source, an inverting converter — are now inside the published ranges, so tolerance analysis on them draws real values.
Connect Claude, Cursor or ChatGPT directly to the tool
Electronics Architect is now an MCP server you can add to an AI assistant by address — paste https://electronics-architect.com/mcp into the client, sign in once, and the tools appear: list the shipped templates, read the design schema, validate and solve a design, run a Monte Carlo, save to the cloud, share and export. The assistant works on the same solver and the same monthly quotas as the editor, and every design it saves lands in My Designs for a person to open and approve. Signing in creates an ordinary API key labelled with the app's name under Account → Developer, where one click revokes it. Pro and Team plans; details at /developers/api.
An API for scripts and AI agents
Pro and Team accounts can now create API keys under Account → Developer and drive the tool from code: validate a design against the published schema, solve it, run a seeded Monte Carlo, save it to the cloud, share it, and export the bill of materials — on the same solver and the same monthly quotas as the editor. Designs an agent saves show up in My Designs, ready for a person to open and review. The reference, with a full worked example in curl, Python and TypeScript, is at /developers/api.
A warning when a supply is too weak to hold up its converter
The simulator now measures how stiff each converter's input supply is and flags the ones sitting close to the point where the design stops working. That includes the coil-and-capacitor filter many designs put on a converter's input, which can be far weaker at its own ringing frequency than its resistance suggests — and it names the damping that would fix it.
Tell the tool what kind of converter you are using
Pick a converter type and the simulator checks whether it can actually make the voltage you asked for — and says so when it cannot.
Mains-powered designs now show the current their rectifier really draws
A rectifier with a smoothing capacitor pulls its current in short, tall pulses rather than a smooth sine. The simulator now works out that pulse for your circuit: the peak current your diodes must survive, the true power factor and distortion your transformer, wiring and fuse must be sized for, and the ripple current the capacitor has to absorb. It also flags when a mains-connected design would fall foul of the European harmonic-current limits.
Team designs now remember who changed what
Anyone on a team with the Editor role can save changes straight back to a shared design. Every save keeps the previous version, records who made it and the note they left, and shows up in the team's activity feed.
Share links that expire on their own
A public link to a design can now be set to run out after a day, a week, a month or three. Send a supplier a power tree for a quote without leaving the door open afterwards. Links you have already made keep working exactly as they do today — expiry is something you choose per link, and the share dialog tells you when an existing one runs out.
Find a real part without leaving your design
Search a part number, or browse the DigiKey catalog, and the brand, package and spec sheet come straight in. We rank live distributor results against the operating conditions already on your sheet, so what floats to the top actually fits, and the filters cut a catalog of millions down to a handful worth reading. It works on both sides of the design too. Look up a microcontroller, FPGA, sensor or memory part and its supply range, running current and sleep current come back filled in, the same way a regulator’s do.
The divider calculator shows its working
Tell it the tolerance you need and it hands back the cheapest standard resistor pair that hits it. Then it breaks down where your error actually comes from. On a typical 12 V to 3.3 V divider the resistors are usually the smallest contributor, well behind supply tolerance and temperature drift. You also get the output impedance your ADC cares about, and if a target isn’t reachable it tells you which change would get you there.
Paste a spec sheet, get a part
Some parts we don’t have yet, and some you can’t share. Paste the text from a spec sheet and we’ll turn it into a component you can use, kept private to your account. If it’s one other engineers would want, a single checkbox offers it up for review.
Know where every number came from
Every spec on a part now shows its origin: checked by hand, confirmed by other engineers, or read off a datasheet by AI. Type over a value yourself and it says that too.
Find out how many will pass
Real parts are never quite as good as their spec sheets. Give us the limits your design has to hit and we’ll estimate what share of units would pass, before you build a board.
Works on tablets
Pinch to zoom, drag to place, tap to inspect. Build or review a design at the bench instead of back at your desk.
Exports worth handing over
Save a design as a crisp image or a print-ready PDF with a proper drawing border. Export your analysis as a report, or bundle the lot into one document to send to a colleague or a client.
Diodes that match their spec sheet
Diodes now behave the way the manufacturer says they will. Pick one from the built-in library, or paste in the numbers from a spec sheet and we’ll work out the rest and tell you how close the match is.
Community-checked part data
If you find a wrong number in our parts database, propose a correction and other users vote on it. Anything still in dispute gets flagged on your design, so bad data can’t quietly slip through.
Share a design, get feedback
Post a working design and ask other engineers what they’d do differently. Replies thread, votes push the best answers up, and the spam gets filtered out for you.
Drag to set efficiency
Shape a converter’s efficiency curve by dragging it, using the same numbers manufacturers publish. The simulation keeps up as you drag, so you see the effect straight away.
Describe it, we’ll draw it
Say what you need in plain English and the AI Generate button builds the design for you. Take it from there yourself, or keep asking for changes.
Spec sheets one click away
Every part you’ve picked links straight to the manufacturer’s spec sheet, right there in the sidebar.
Zener diodes
Zener diodes simulate properly now, with a breakdown voltage you can set and a graph showing how they behave in both directions.
Start at the wall socket
A design no longer has to begin at a DC rail. Transformers, rectifiers, inverters and motors, single or three-phase, sit alongside your DC circuits in the same drawing.