Sasha
What Sasha is
Sasha is a platform for teams that run Claude on their own infrastructure. Conversations and documents stay inside your environment, and the team can share the prompts and knowledge that work. Sasha also records how each prompt performs.
Help and recipes
Worked examples and help for Sasha, by topic.
- Find and answer Find what your organisation already knows and get answers with sources.
- Meetings Prepare for meetings and keep a record of what was decided.
- Bring knowledge in Add documents and drives so Sasha can use them.
- Write and publish Write new documents from what you have, and publish facts to your website. Coming soon
- Team and admin See who is doing what, and manage people, access, skills and apps.
- Use Sasha from Claude Connect Claude to your Sasha and work from there.
- Private and secure Where your knowledge lives and who can reach it.
- Product guide How each part of Sasha works, screen by screen, for admins.
- For AI agents and developers How to use every Sasha MCP tool, prompt and resource.
Run Claude on your infrastructure
Sasha runs Claude through the cloud account you already pay for.
Providers:
- AWS Bedrock. Use your existing AWS agreement. AI calls route through your VPC under your security policies.
- Google Vertex AI. Use your GCP infrastructure and the compliance certifications you already hold.
- Anthropic API. Direct access with no cloud intermediary.
Effects:
- AI spending sits under your existing cloud contract.
- Your security team has already approved the infrastructure.
- No new vendor relationship or procurement cycle.
- You can switch providers without rewriting prompts or retraining the team.
You control the model version, the context window and the rate limits. You change providers in settings.
Measure prompt quality
Sasha records what each prompt produces and shows you which prompts work.
Prompt effectiveness:
- Quality scores for each conversation.
- A view of which prompts produce accurate responses and which produce confusion.
- Response quality tied to specific CLAUDE.md versions.
Hallucination detection:
- Flags responses that do not match your knowledge base.
- Surfaces sessions that need human review.
- Builds confidence scores from source attribution.
A/B testing:
- Test prompt variations against each other.
- Compare system prompt versions with measured outcomes.
- Roll back to a previous version when needed.
Quality dashboard:
- Sessions that need attention.
- User activity and engagement patterns.
- Error analysis and categorisation.
- Resource usage and cost tracking.
Shared prompts and knowledge
A prompt that works for one person is available to the whole team.
Prompt library:
- Star and share prompts.
- Click to insert a saved prompt.
- Organise prompts by category, project or use case.
- Version history for each prompt.
Shared knowledge base:
- One documentation store for the whole team.
- Nested folders that follow your team structure.
- Updates appear for all users without email attachments.
User activity:
- See how the team uses Sasha.
- Identify frequent users and the techniques they use.
- Spot adoption gaps.
Connect your corporate data
Your documents stay where they are. Sasha connects to them.
Integrations:
- SharePoint and OneDrive. Access your Microsoft 365 document libraries.
- Google Drive. Connect shared drives and team folders.
- Network File Systems. Mount remote corporate storage.
- Local Repositories. Git repositories and local folders.
No migration:
- Documents remain in their current locations.
- Existing permissions and access controls apply.
- Sasha does not copy sensitive files to a third-party system.
- Search runs across all connected sources.
Access security:
- OAuth authentication for cloud providers.
- Service account support for automated access.
- Sasha encrypts credentials at rest.
- An audit log entry for each file access.
Privacy and isolation
Share-nothing architecture:
- Each deployment is isolated. There is no shared infrastructure.
- Your instance runs in your environment or in a dedicated container.
- No cross-tenant data exposure.
Your data stays in place:
- Sasha stores conversations in your database.
- Documents do not leave your network for processing.
- Your content is not used for training.
- One command deletes all data.
Audit trail:
- Full conversation history with timestamps.
- Sasha logs permission grants and denials.
- Sasha tracks hook executions with user attribution.
- You can export audit trails for compliance review.
Compliance:
- HIPAA-compatible deployment options.
- GDPR data residency controls.
- SOX audit trail requirements supported.
- Your security team reviews the architecture.
Get started
Browse the guides in the sidebar, or start a conversation.