Product

One workspace from analysis code to signed document.

Writers, statisticians, reviewers and agents work from the same verified definitions and values. When the analysis changes, every document shows which numbers moved.

Book a pilot Reproduce from code

Evidence tab: the analysis cohort as steps in code order, each with its patient count and a link to the lines that apply it.

The cohort, read from the code

Each inclusion and exclusion step shows the patients left after it and links to the lines that apply it. CDISC ADaM names read in plain English.

SQL, R and Python
Analysts verify each definition
Verification goes stale when the code changes

Document workspace with tracked insertions and deletions and a panel to accept or reject each suggestion.

Live numbers, tracked changes

Writers, co-authors and agents work in one draft. When the analysis changes and the document syncs, linked numbers update and Silogram shows what moved; the number audit finds numbers typed by hand.

Word export with real tracked changes
Import a co-author's edits as suggestions
Journal requirements, each with its source

House style learned from past documents: spelling, voice, number formats, preferred terms and abbreviations.

Your house style, learned from past papers

Upload past manuscripts and abstracts. Silogram learns spelling, voice, how you write P values and CIs, preferred terms and abbreviations.

Every draft and every agent follows it
Mechanical fixes arrive as tracked suggestions
Never used to train AI models

Connected agents with their scopes, and the form that gives a new agent scoped, revocable access.

Agents suggest. People decide.

The writing assistant runs on the model your company has approved. Outside agents connect over MCP or the REST API with scoped, revocable tokens.

Edits arrive as tracked suggestions under the agent's name
Numbers they write are linked or flagged
Every call logged for AI-use disclosure

Evidence-to-access card: each decision waiting on a document, its likely and 90% dates and its value per week (synthetic data).

Forecasts tied to the decisions they unlock

Most-likely and 90% dates for every document, recalculated from real progress, and linked to the coverage, formulary, guideline or approval decision it supports.

See which phase is slipping
Each week priced with your own revenue estimate

Speed to access

Sign-off tab: author, review and approval signatures bound to the document's digest, with the document locked after approval.

Signed, locked and on the record

Signatures record the signer, time and meaning, bound to a digest of the exact text. Approval locks the document until someone reopens it with a reason.

Security and compliance

New · Reproduce from code

Proof that the code produces the numbers.

A runner re-runs your analysis at a commit, in a container with no network access, and Silogram checks that it produces every value, table and figure your documents cite.

Your runner sits inside your network, next to the data; only the declared outputs and a status line come back
Pin it to your own repository and it runs only committed code with the committed command
Or use Silogram's hosted sandbox for synthetic or de-identified data
Read-only data, no network, no privileges, time and memory limits
Each run, its log and the comparison are in the audit trail

Reproduce from code: all 12 values, 1 table and 1 figure of a synthetic validation analysis match the committed outputs.

Publications, clinical, regulatory and market-access documents.

Each document type gets its own outline, plan, drafting voice and check. All of them draw on the same verified analysis.

Publications

Journal manuscript

ICMJE with STROBE, RECORD, CONSORT 2025, STARD 2015 or TRIPOD+AI; the target journal's requirements, each with its source.

Publications

Congress abstract

Background, methods, results and conclusions, with a plan forecast against the submission deadline.

Clinical

Clinical study report

ICH E3 outline from the synopsis to overall conclusions; drafted from the trial's outputs with document QC.

Clinical

Protocol and SAP

ICH M11-style protocols and SAPs to ICH E9(R1), with every analysis population named.

Regulatory

Clinical summaries

CTD Module 2 efficacy and safety summaries, checked against the CSRs.

Market access

Payer and HTA dossiers

AMCP Format 5.0 formulary dossiers, EU joint clinical assessment dossiers and MolDX technical assessment packages.

Every outline includes a use-of-AI section, drafted from the log of model and agent calls.

Connect the agents you already use.

An agent's edit reaches the text only after a person accepts it. Tokens are shown once, stored as a hash, rate-limited and revocable.

read: verified definitions, values, sections and house style
suggest: edits that arrive as tracked suggestions
comment: notes on a section for the writers

Add Silogram to any MCP client

// Token from Settings › AI & agents
{
	"mcpServers": {
		"silogram": {
			"type": "http",
			"url": "https://app.silogram.ai/api/mcp",
			"headers": {
				"Authorization": "Bearer sg_…"
			}
		}
	}
}