Skip to content

Capabilities

This page is the complete catalog of what Myra AI Workspace can do. Capabilities are grouped by what they let you accomplish; each entry links to its detailed page. For the high-level overview and architecture, see What is Myra AI Workspace.

Chat and everyday use

  • Chat — Persistent, multi-turn chat with a model picker (including Auto), conversation history, projects, and exports. The everyday way to work with AI through the gateway.
  • Attaching files to a message — Send a document, spreadsheet, or image with your message and ask questions about it.
  • Read a reply aloud — Have any finished answer spoken with an EU-hosted voice (text-to-speech).
  • Recording a voice message — Dictate your message instead of typing it (speech-to-text).
  • Follow-up suggestions — One-click prompt chips that continue the conversation naturally.
  • Memory — The model remembers facts, preferences, and instructions across conversations; suggested memories are yours to keep, edit, or reject.
  • Personal style and commands — A saved writing style, and reusable /name prompt templates with variables, categories, search, versioning, and sharing.
  • Prompts (Library) — An organisation-wide catalog of admin-managed prompt templates (the Prompts tab of the Workspace Library group), versioned with full snapshot restore, categories, full-text search, and a Use in chat row action — the org-level counterpart to personal commands.
  • Web search — Answers grounded in live web results, with inline citations, a freshness picker, and a per-step process timeline; the search query egresses only after PII is tokenised.
  • Extended thinking — A collapsible Thinking block shows the model's reasoning and how long it took, on models that support it.
  • Archiving conversations — Rename, archive, and reveal archived conversations to keep the history list focused.
  • Shared conversations — Public share links and live multi-user rooms.
  • Ghost mode — A conversation kept in memory only: nothing is saved or logged.
  • Automatic context management — Long conversations are compacted automatically, in the background, so they never hit a wall.

Knowledge, documents, and OCR

  • Knowledge files (with OCR) — Upload documents, spreadsheets, presentations, and structured data — and scanned PDFs, photos, and images. Text is extracted with optical character recognition (OCR) and made searchable and citable in chat. Supported formats include PDF, DOCX, ODT, XLSX, PPTX, CSV, TXT, Markdown, XML, JSON, HTML, and PNG/JPG/JPEG/WEBP images.
  • Add a website as a source — Pull a public web page into a project's knowledge and keep it current.
  • Connected sources & sync — Sync knowledge from SharePoint, Confluence, or a website, with automatic re-crawling and permission mirroring. See also Supported data-source types.
  • Data-file answers — Spreadsheet and CSV answers lead with exact row, column, and sheet counts, and flag when a value is estimated rather than computed.
  • Long and data-file turns — The gateway bounds large re-sent previews and automatically upgrades a data-file turn to a capable model so the file can actually be processed.
  • Conversation file space — Even outside a project, every conversation owns a file space where uploads and generated files persist for the rest of the thread.

Projects and collaboration

  • Projects — Workspaces that group conversations under shared instructions, a knowledge corpus, a pinned model, and members with roles.
  • Project members and roles — Owner, editor, and viewer access, grantable directly or through Groups.
  • Groups — Organise users into groups and grant a whole group scoped access to knowledge spaces, projects, and agents at once.
  • Project access tiers — Set a project to local-only, PII-protection-required, or user-configurable, controlling which models are available and whether masking is forced.
  • Sharing knowledge org-wide — Publish a synced source to the whole organisation.
  • Comments on a reply — Discuss a single answer with project members.

Agents

  • Agents — Reusable assistants that combine a system prompt, a model, tools, and bound knowledge, run from chat, on a schedule, or from a webhook.
  • Try-It preview — Run a draft agent before saving it.
  • Versions and organisation catalog — Numbered versions with restore and side-by-side comparison, and a publication-governed catalog a KI-manager approves before every member can clone from it.
  • Chat bridge — Talk to an agent from Mattermost, Microsoft Teams, or Slack, including native /agents and /agent slash commands in Mattermost, with bot-token rotation and an opt-in for web search while all other egress stays blocked.
  • Webhook triggers — Fire an agent or workflow run from an external system through a secret-protected inbound URL, optionally under a no-egress tool profile that strips every data-exfiltrating tool.
  • Agent approvals (egress review) — Hold a scheduled agent run's external delivery until an administrator approves it; denied or expired holds are never sent.

Workflows and automation

  • Workflow builder — Chain agents and delivery steps into one automated run, open to every authoring role, and test a single step against sample input before publishing.
  • Runs monitoring — A runs list and per-run detail with per-step status, cost, and PII state, and the ability to cancel a run in flight.
  • Triggers — Start a workflow manually, on a schedule, from a webhook, or from a public form.
  • Public forms with file upload — Collect input (and files) from people without an account, with a live PII check before anything is persisted.
  • Human-approval steps and the Approvals — Personal tab — Pause a run for a nominated approver; the Personal tab of Settings › Approvals collects the approvals assigned to you.
  • Scheduled tasks — Run a prompt on a timer with results delivered into chat and optionally by email, with failure notifications and auto-pause.

Data analysis and generation

  • Code interpreter — A sandbox runs Python on an attached file, produces charts, and returns downloadable results — on EU and open models, not only Anthropic.
  • Image generation — Generate and edit images from a prompt on an EU-hosted model.
  • Artifacts — Preview and edit generated documents and rich text inline; generated files appear as artifact cards and persist to the conversation's file list, exportable as Office or OpenDocument formats (.docx, .xlsx, .pptx, .odt, .ods, .odp).
  • Playground — Compare several models side by side on the same prompt.

Governance and compliance

  • Governance templates — Map EU AI Act and GDPR obligations onto gateway configuration, preview the diff, and apply it — never below your data-residency floor.
  • Governance dashboard — One tenant-scoped view of guardrail activity, residency split, templates, pending approvals, and compliance-report history.
  • Compliance reports — Scheduled monthly snapshots and on-demand exports as CSV or PDF.
  • Tamper-evident audit log — A cryptographic hash-chain seals every audit record (revisionssicher) so tampering, gaps, or truncation are detectable.
  • Config approvals (four-eyes) — Hold a prompt-library or agent change until a second, different administrator approves it.
  • Retention and deletion — Per-tenant, per-project, and per-user retention limits (Löschfristen), right-to-erasure, and a machine-readable deletion protocol.
  • Self-service data-subject rights — Any user can export a machine-readable copy of their own data (GDPR Art. 15) and deactivate or irreversibly erase their account (Art. 17); an administrator can place a reversible processing restriction on a user (Art. 18).
  • Contract-end export — Hand over all tenant data in open formats (JSON and CSV) at the end of a contract.
  • AI-generated labelling — Every answer is marked AI-generated, as required by the EU AI Act.

Security and privacy

Access, identity, and administration

  • Multi-tenancy — Isolated tenants, each with its own gateways, keys, tokens, rate limits, and routing — nothing crosses the boundary.
  • Gateways — Named deployments within a tenant (e.g. production, staging), each the configuration surface for its own guardrails, routing, budgets, providers, and keys.
  • Single sign-on (OIDC & SAML) — Redirect to your identity provider, with automatic login-page domain detection and optional department-to-group mapping from an IdP attribute, alongside email one-time-code sign-in and 8-hour or 30-day sessions.
  • User management and roles — Create, edit, role (admin, tenant admin, KI-manager, member, finance, viewer, demouser), and reassign accounts between tenants, searchable and exportable as CSV or XLSX.
  • Access tokens — Scoped myra_ inference tokens with expiry, spend caps, and per-token rate limits, hashed at rest; users manage their own tokens without an administrator.
  • Organisation policy — A tenant-wide instruction added automatically to every chat and agent interaction across the organisation.
  • White-label branding — Per-tenant product name, light/dark logos, favicon, brand colour and font, home greeting, chat disclaimer, and sidebar links, applied at runtime across the app, login pages, transactional emails, and document exports.

Providers and routing

  • 21 provider integrations — OpenAI, Anthropic, Google Gemini and Vertex AI, AWS Bedrock, Azure OpenAI, Mistral, Cohere, HuggingFace, Cloudflare Workers AI, Myra's EU fleet, and more.
  • Unified compat endpoint — A drop-in OpenAI-compatible endpoint that resolves the provider automatically from the model name, translating OpenAI tool_calls to native Anthropic tool_use and back.
  • Provider-key rotation and failover — Self-serve tenants automatically fail over from a blocked or exhausted provider key to a healthy standby key from the pool.
  • Routing rules — Rewrite provider and model, and split traffic by weight, based on request metadata.
  • Fallback and retry, load balancing, and circuit breaker — Keep traffic flowing when a provider fails or slows.
  • Response caching — Serve repeated identical prompts from cache to save cost and latency.

Cost, budgets, and rate limits

  • Budgets — Hard spending caps per token, per gateway, and per tenant, with automatic period resets or a non-resetting total lifetime period.
  • Budget alerts — In-app banners when any budget nears or reaches its cap, plus per-token consumption visibility; self-serve tenants also get allowance emails at 80% and 100%.
  • Provider Costs — The per-model pricing table (Settings › Costs › Provider Costs) that drives cost attribution, with a sync from upstream sources.
  • Rate limiting — Sliding-window request caps at the gateway and per token.

Activity and analytics

Integrations and connectors

Sign-up, billing, and API

See also