Capabilities¶
What Agent-Nuvira can do beyond "write a file". Every count here is produced by a command you can run yourself.
| Surface | Size | Command |
|---|---|---|
| Agent tools | 110 | nuvira tools list |
| CLI command entries | 289 across 48 groups | see Full Command Surface |
| Inference providers | 22 | nuvira provider list |
| Messaging platforms | 22 | nuvira gateway status |
| Skills | 155 on disk | nuvira skills list |
| Workflow templates | 10 built in | nuvira workflow list |
| Dashboard pages | 22 | see the User Manual |
Coding¶
Plan, implement, review, test and publish. nuvira execute runs a multi-agent pipeline;
nuvira workflow run runs a fixed one; nuvira chat is the interactive surface. Every
mode shares one tool surface and one router.
Beyond coding¶
- Browser automation — drive a real page: open, click, type, extract, screenshot.
- Vision — describe screenshots, diagrams and photos via a local Ollama model or a free vision tier.
- Image and video generation — from a prompt, saved to the artifact directory.
- Voice in both directions — synthesize speech and transcribe audio locally.
- Desktop control — screenshot, mouse and keyboard without stealing focus.
- Sandboxed code execution — run code under time and resource limits.
- Document extraction — PDF, DOCX, XLSX, PPTX, HTML, CSV.
- Assessment of other repositories — shallow-clone a repo into an ephemeral cache and work on it without touching your own workspace.
- Integrations — Docker, Microsoft Graph (mail/calendar/OneDrive), Home Assistant, Discord, Feishu docs and drive, kanban boards.
- Scheduling and supervision — cron jobs and long-lived daemons under the agent's control.
- Sub-agents — delegate a focused subtask to a fresh, isolated context and get the summary back.
Messaging¶
Twenty-two platforms, from Telegram and Slack to SMS, IRC and Home Assistant. Inbound messages are dispatched through the same pipeline as the CLI, and progress plus the final result are sent back to the channel. Two independent permission lists govern who may trigger the agent and who may direct it to send to other people.
Knowledge and memory¶
Trajectory memory, fact memory, coding-pattern extraction and failure lessons. Vector retrieval keeps long contexts affordable — the reference machine measured 142,493 tokens saved at a 65.6% average reduction across 64 calls.
Extension¶
- Skills — reusable, parameterised execution plans.
- Workflows — fixed pipelines you can search, install, publish.
- MCP — connect external tool servers, or expose the agent's own tools as an MCP server.
- Plugins — providers, agents and workflow templates.
- SDK — build custom agents.
- Federation — connect remote agent instances, including A2A.
Verification and governance¶
Per-step traces, an evaluation framework, model benchmarks, a CycloneDX software bill of materials, security scanning for injection and PII, an admin governance policy (allow/deny providers and models, hard cost cap, privacy floor), and RBAC.
Try it without spending a token¶
nuvira doctor # full environment diagnosis
nuvira tools list # the 110 tools, with descriptions
nuvira skills search docker # find a skill by need
nuvira skill run security-audit --dry-run # print the plan, run nothing
nuvira workflow list # the ten built-in pipelines
nuvira intent resolve "stop the dashboard" # plain English -> exact command
nuvira code-map src # symbol map of a directory
nuvira retrieval stats # measured token savings