SPEAKMAN.AI · Release v1.3.0

Rendering a document no longer means installing anything, not on your machine and not inside whatever's calling the tool.

By David Speakman ·

SPEAKMAN.AI is the free, local, open source engine behind the same MCP workflows the hosted platform runs. Version 1.3.0 moves document rendering onto the MCP server itself, so no client needs a local script to generate a SAD or compliance report. That's the change that let generate-sad run end-to-end inside Microsoft Copilot Studio for the first time. Four engine bugs this release's own HITL testing caught in the act are fixed too.

speakmanai.log — real output, this release INFO Processing request of type CallToolRequest INFO Calling render_document(markdown, title, subtitle) INFO render_document -> 200 OK, self-contained HTML returned

That call is the whole rendering step now. No local script runs, no Node subprocess spins up, and the default branding no longer costs a base64 logo in the response.

~1,400
tokens for a full rendered document, down from ~57,000 with an unoptimized embedded logo.
0
local dependencies left for document rendering. No Python, no Node, no puppeteer.
1
new MCP client confirmed end-to-end: Microsoft Copilot Studio.
4
engine bugs found and fixed, surfaced by this release's own HITL review gates.
Why This Release

Generating a document used to mean shelling out. Now it's one tool call.

generate-sad and generate-compliance-report used to finish by running a local script to build HTML, then attempting a PDF through a Node puppeteer subprocess. That's fine when the calling client is Claude Code with a full shell. It's a hard stop for a client whose skill runtime can't shell out to anything at all. That's most clients, once you stop assuming Claude Code specifically.

render_document moves the whole step onto the MCP server. Any client sends Markdown, gets a fully self-contained HTML document back, and never touches a local interpreter. PDF is gone along with it. The HTML is print-to-PDF-able from any browser, so nothing was actually lost by dropping a Node dependency nobody wanted to manage.

Confirmed, Not Assumed

The protocol was never Anthropic-only. The skills were, until Copilot Studio.

SPEAKMAN.AI's own hosted UI already talks MCP instead of a bespoke API, and the engine has been run from other AI tools directly. What hadn't been tested until now was the skills layer itself: the packaging built for Claude Code and Claude Desktop, running from a client outside that ecosystem for the first time. generate-sad ran to completion inside a live Microsoft Copilot Studio agent, connected to a local SPEAKMAN.AI server over a tunneled connection, MCP connector attached. The only thing that changed from a Claude Code run of the same skill was which tool called it.

Before

A skill runtime that can't shell out to Python or Node can't run generate-sad at all.

After

Same skill, one MCP tool call for rendering. Verified working end-to-end in Copilot Studio this release.

What Shipped

The full list, not just the headline.

Rendering

render_document: server-side Markdown-to-HTML, no client-side dependency. Default branding is a plain text wordmark rather than an embedded image; the token-cost difference is the stat above.

Execution

Fixed a step declaring more than one agent silently running only the first, a latent bug present since the engine's original design and confirmed absent from the enterprise version.

Review

Fixed the HITL loop cap discarding a human's final round of feedback and advancing on stale, pre-feedback content instead of what was actually just regenerated.

Coverage

RE_USE_CASE_ANALYST_V1 previously produced one use case per requirement by construction. It now enumerates every use case a requirement actually implies, roughly 5x the catalog density on a real test project.

Where This Sits

One engine. Free or hosted, it runs the same pipeline.

The hosted SPEAKMAN.AI platform runs a full idea-to-production SDLC pipeline. This repository is the free, local, MIT licensed engine that executes the same MCP workflows, with full functionality for an individual user and no account required. As of this release, there's also no assumption that the calling client is Claude Code.

Idea Intake

/generate-project-brief

A raw idea, coached into a structured brief. No SPEAKMAN.AI session, no credits.

Architecture

/generate-sad This release

Business description in, Solution Architecture Document out. Now client-agnostic end to end.

Requirements

/generate-requirements This release

Domain model, naming dictionary, use cases, API and DB schema. Agents decomposed and use-case coverage fixed this release.

UX Design

/generate-ux-design

Screen inventory, flow diagrams, a proposed direction, and HTML mockups for the highest-priority screens.

Code Generation

/generate-speakmanai-code

A working, milestone gated, git committed codebase.

Infrastructure

/generate-infrastructure

Terraform for GCP, AWS, and Azure, plus pre-pentest hardening.

Pentest & UAT

/pentest

A gated security engagement, run against the live staging build.

Runs On Your Machine, From Whatever Talks MCP

The free tier is not a trial. It is the whole engine.

ModeCommandDatabase
Windows exeRun SpeakmanAI.exe~/.speakmanai/speakmanai.db
Docker + SQLitedocker-compose up --buildspeakmanai_data volume
Docker + MongoDBdocker-compose --profile mongo upMongoDB, port 27017
Dev, no Dockeruvicorn server:app --port 8000~/.speakmanai/speakmanai.db

There's no account and no cloud dependency. No per-call fee beyond your own provider key, either. The Windows build is now code-signed, so installing it doesn't come with a SmartScreen warning telling you to turn back.

Individual · Free · MIT

Run the engine locally.

Download the signed exe, or clone and run with Docker. You get full functionality on your own API keys, with nothing sent anywhere else.

Teams · Hosted

Need multi-tenant support and audit trails? Metered billing too?

The hosted SPEAKMAN.AI platform runs the same engine with the infrastructure teams need on top of it.

Written by David Speakman. Speakman Consulting designs and builds this kind of system for growing organizations: agent workflows with the governance that keeps humans in the loop.