Research & Chat
Deep research,every claim cited.
Ask Sofie a hard question and it researches across your workspace files, regulatory sources and the web. Every statement links back to where it came from.
HT-7 hold-time support
Deep researchSources
6 of 18 citedHTS-014 Hold-Time Study Report.pdf
p. 14 · exact
21 CFR 211.111 · Time limitations on production
ecfr.gov
Process Validation: General Principles and Practices
fda.gov · Guidance
LB-204 Transfer Protocol v3.docx
§6.2 · exact
BR-2291 Executed Batch Record.pdf
p. 22 · exact
Designing hold-time studies for biologics intermediates
bioprocessintl.com
Inregulatedwork,ananswerwithoutasourceisaliability.Sofieshowsyouwhereeachclaimcamefrom,andtellsyouplainlywhentheevidenceisn'tthere.
Capabilities
What Research & Chat does
01
Deep research
Sources drawn from your workspace, nine regulatory authorities and the web, with every claim cited.
02
Citations you can check
Inline citation bubbles and a sources panel with the passage behind each claim. Missing evidence is named, not filled in.
03
Sixteen models and Auto
Models from Anthropic, OpenAI, Google, Meta, xAI, Moonshot AI and Z.ai. Auto routes each request to a fast, balanced or deep model.
04
Plan mode
Turn on Plan Mode and Sofie lays out a step-by-step plan before it acts. Adjust it in the chat, then execute.
05
Edit and branch
Edit an earlier message, or branch the conversation into a new chat to try another approach. Search past conversations and keep long chats summarized.
06
File Studio and Sofie Guide
Edit PowerPoint files in File Studio, generate images and interactive visuals, and ask Sofie Guide how anything in the platform works.
Sofie Everywhere
On every page, one question away.
The Sofie sidebar stays with you across the platform. Ask why an orchestration pauses where it does, what changed in a tracker, or how to set something up, without leaving the page you are on.
Orchestration
Sofie
The QA review point holds the run until a reviewer confirms the deviation list before the summary is drafted.
Models
Every leading model. One place to use them.
Choose a model for the task, or leave it on Auto. Auto sends quick lookups to a fast model and hard reasoning to a deep one, and its life-sciences safety minimums send consequential work, such as regulatory or quality judgments, to a deep model.
- Anthropic Claude, OpenAI GPT and Google Gemini
- Meta Muse Spark, xAI Grok, Kimi and GLM
- Auto and the leading models also power orchestration agents
Auto
Routes each request to a fast, balanced or deep model, with life-sciences safety minimums for consequential work.
Anthropic
- Claude Fable 5.1
- Claude Opus 5.5
- Claude Sonnet 5.5
- Claude Haiku 4.5
OpenAI
- GPT-6 Astra
- GPT-6.1 Sol
- GPT-5.6 Terra
- GPT-6 Luna
- Gemini 3.1 Pro
- Gemini 3.8 Flash
- Gemini 3.5 Flash Lite
Meta
- Muse Spark 1.2
xAI
- Grok 4.6
Moonshot AI
- Kimi K2.6
Z.ai
- GLM 5.3
- GLM 5.3 Flash
Plan and branch
Agree the plan. Then execute it.
With Plan Mode on, see how Sofie intends to approach a gap assessment across forty documents before it starts. Change the scope in the chat, then execute. For a second angle, branch the conversation into a new chat.
Plan
Plan Mode- 1.Read the transfer protocol and the sending site's hold-time studies
- 2.Pull the receiving site's equipment list and cleaning SOPs
- 3.Compare hold times and cleaning limits against the protocol
- 4.Draft a gap summary with citations as a CoDraft
From answer to deliverable
Open the answer as a document.
When the research is right, open it as a CoDraft with the citations carried through, and keep working with your team in a Word-faithful editor.
Tech Transfer Charter · Controlled
LB-204 Drug Substance Transfer
Version 0.4 · Draft for QA review
3. Scope of Transfer
This charter governs the transfer of the LB-204 upstream and downstream process from Larkspur Biologics, Cambridge (sending site) to {Receiving site} (receiving site). The transfer covers the 2,000 L fed-batch process through drug substance release, including fill-finish operations.
4. Roles and Responsibilities
5. Process Overview
“What used to take me half a day searching through guidances and cross-referencing documents, Sofie handled in seconds. The answer was there; the references were there; nothing left to chase down.”
Questions
Frequently asked
Which models can we use?
Sixteen models: Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5; GPT-6 Astra, GPT-6.1 Sol, GPT-5.6 Terra and GPT-6 Luna; Gemini 3.1 Pro, 3.8 Flash and 3.5 Flash Lite; Muse Spark 1.2; Grok 4.6; Kimi K2.6; and GLM 5.3 and 5.3 Flash. Auto is the default.
How does Auto choose a model?
Auto routes each request to a fast, balanced or deep model based on what it needs, and its life-sciences safety minimums send consequential work to a deep model. Choose a specific model whenever you prefer.
What does Sofie do when it can't find a source?
It says so. Sofie names the gap rather than filling it from memory, so you know what still needs evidence.
Is our data used to train models?
No. Customer content is used to do your work, and no model is trained on it.
Works with
Part of one connected platform
Regulatory Intelligence
Nine authorities, cited and monitored every day.
CoDraft
Word-faithful documents Sofie drafts, redlines and cites with your team.
Memory
Institutional context that carries forward, linked to its evidence.
Sofie Voice
Talk through the work while Sofie builds it beside you in split screen.
Start with one workflow.
Measure it. Then expand.
In a working session we choose a priority workflow, define your standards and review points, and agree how to measure speed, quality and capacity. Then you evaluate Sofie's work before deciding whether to expand.