BetterCallClaude.ch 4.8.3: Case Law Goes Interactive, the Pipeline Goes Parallel

From 4.7.0 to 4.8.3: local playbooks, native widgets inside the conversation, verifiable deadline computation, and legal research running in parallel. The story of three releases.

In January, Anthropic did two things that matter to anyone building legal tools on Claude: it launched Cowork with a suite of official plugins — including a Legal plugin — and it published MCP Apps, the first official extension to the Model Context Protocol, which lets servers return interactive interfaces directly inside the conversation.

The official Legal plugin is a good product, with one stated limitation: its playbooks are written for Delaware, New York, and California. If you practice under any other legal system, you have to adapt it yourself.

We already have that legal system. The releases from 4.7.0 to 4.8.3 of BetterCallClaude.ch start exactly there: Anthropic validated the category — we own the jurisdiction.

Your firm's playbook: bettercallclaude.local.md

The most requested feature from teams using BCC in-house: a local playbook file where the firm — or the Rechtsdienst — encodes its standard positions. Preferred governing law, forum, acceptable liability caps, limits for Konventionalstrafen, escalation thresholds, working language, and citation style. The template ships in German, French, Italian, and English, and the language you fill it in drives the language of your deliverables.

From that moment on, every contract analysis compares clauses against your positions, not generic ones. And there is one rule that is not up for negotiation: the playbook customizes preferences and thresholds, but it can never override mandatory law. If a configured position conflicts with non-derogable rules — think Art. 100 CO — BCC flags it instead of silently applying it. That is the safety net a playbook designed for American law cannot give you.

If you are coming from the official Legal plugin, migration costs nothing: BCC also reads Anthropic's legal.local.md format, tells you which US-centric positions are being ignored, and offers to convert.

NDA triage, Swiss style

Built on top of the playbook is the new NDA triage command, with the traffic-light classification the official plugin made familiar — but with Swiss-law criteria: forum and the Lugano Convention, Konventionalstrafen within your configured thresholds, entanglement with non-compete restrictions under Art. 340 ff. CO, FADP issues. It also works in batch: a folder of NDAs in Cowork, one summary table, and the individual reports.

Cowork as a first-class environment

BCC was born inside Claude Code, among lawyers who are not afraid of a terminal. With Cowork the audience changed, and 4.8.0 takes that seriously, all the way down:

/bettercallclaude:start welcomes the new user, checks server connectivity, and if no playbook exists, builds one through five plain-language questions. The old setup lives on as an alias and will point the way until v5.0.
/bettercallclaude:doctor diagnoses server status in a lawyer's language, not a sysadmin's: "without this server, citations are not verified."
Deliverables as files, always. Every substantial output lands in an orderly structure (bcc-output/), numbered along the pipeline phases and — above all — accompanied by a sources.md that records every source consulted, with the date of verification. That is not a technical detail: it is the paper trail your firm's due diligence expects from any tool that enters the workflow.

The widgets: case-law research inside the conversation

This is the visible leap. Thanks to MCP Apps, BCC's remote servers no longer return only text:

Federal and cantonal case-law search. A query against bge-search or entscheidsuche opens a results browser inside the chat: filters by chamber, period, language, and canton; the key considerations one click away; and a "Use in analysis" button that carries the selected decision straight into the memo you are working on.

Adversarial dashboard. The three-agent analysis — advocate, adversary, judicial synthesis — becomes a panel with the probability score at a glance and expandable arguments, each with its legal basis and clickable cited decisions.

Citation validation. Paste a brief: the citations come back classified green, yellow, and red, with proposed corrections you can apply with a click — one at a time or all at once.

Intake form. The structured briefing that used to take several rounds of questions in chat is now a single form you fill in under a minute. The adaptive intelligence stays exactly where it was — in the skill that decides which questions to ask — but filling them in stops being a ping-pong match.

One design choice we did not compromise on: the widgets load nothing from third parties. No CDNs, no remote fonts, no analytics. Everything is served from our own servers. When data covered by attorney-client privilege flows through an interface, every outbound network request is one request too many. And because MCP Apps is an open standard, co-developed by Anthropic and OpenAI, these widgets are not tied to a single client: they reach anyone who connects the remote servers, with or without the plugin, and degrade gracefully to text on clients that do not support them.

The deadline calculator: the feature we almost decided not to ship

compute_deadlines calculates ZPO and BGG deadlines — judicial recess suspensions under Art. 145 CPC, computation under Art. 142 ff., public holidays across all 26 cantons — and lays them out on a timeline with the computation expanded step by step.

We will say it plainly: this feature stayed in the drawer until it passed a severe gate, because a miscalculated deadline is the one mistake a lawyer never forgives. Three non-negotiable conditions: a suite of computation cases with known, verifiable outcomes that must pass at one hundred percent; a computation that is never a black box — every day, every suspension, every holiday encountered is made explicit and human-verifiable; and the explicit refusal of out-of-scope combinations rather than a guessed result. The disclaimer is permanent and cannot be dismissed: this is a computation aid, not legal advice, and professional verification remains mandatory. StPO and VwVG will arrive only once they have passed the same gate.

4.8.3: the pipeline learns to run in parallel

The release that gives this announcement its number is also the most experimental one. Claude Code's dynamic workflows (research preview) make it possible to orchestrate multiple subagents in parallel, and the /legal-5step pipeline has two phases that ask for it by nature.

With the --parallel flag, the research phase stops querying sources one after another: four subagents work simultaneously — federal case law, cantonal case law, statutory law on Fedlex, doctrine and commentary — and a synthesis step deduplicates, detects conflicts between sources, and verifies every citation before the memo is written. The paper trail stays complete: sources.md also records who consulted what.

And the adversarial phase gains something beyond speed: advocate and adversary start together, on the exact same dossier, and neither sees the other before the judge's synthesis. In the sequential flow, the adversary worked after the advocate; parallelism eliminates any possible anchoring by construction. Sometimes engineering improves methodology.

All of this is opt-in, explicitly experimental, and degrades silently to the sequential pipeline wherever the preview is not available. The primary path remains the battle-tested one.

Under the hood

The 4.8.x line is also a release about discipline. We consolidated the plugin's skills following Anthropic's guidance on "context tax" — fewer permanent descriptions in context, more knowledge loaded on demand — under one iron rule: every merge had to pass the full evaluation framework (197 test cases) with zero regressions, or it was rolled back. No capability was lost; only where it lives has changed.

And every skill that depends on the servers now explicitly declares how it behaves when a server does not respond. A tool you use for the law must be honest about its own limits, too.

Try it

BetterCallClaude.ch is open source, governed by a Swiss non-profit association, built on open standards — MCP now sits under Linux Foundation governance — and on the public data of Swiss justice: BGE, Fedlex, entscheidsuche, OnlineKommentar.

BetterCallClaude.ch the Swiss edition has been officially approved and recognised by Anthropic.

You will find it in the Claude plugin marketplace at https://github.com/fedec65/bettercallclaude; the free tutorial is available here: https://github.com/fedec65/bettercallclaude_TUTORIAL; the remote servers at https://github.com/fedec65/bettercallclaudeMCP also work standalone. The fastest way to see what changed: install it, type /bettercallclaude:start, and search for a decision.

**** IMPORTANT ****

BetterCallClaude does not replace the advice of a licensed attorney. It is a working tool for legal professionals — and as of this release, a tool you work with inside the conversation, not just through it.