Documentation

What's inside:

• Agent Studio, create and manage agents with custom prompts, models, and tool access
• Workspace, chat with agents directly in the browser
• Channels, deploy agents to Slack, WhatsApp, Telegram, Discord, WeChat, and DingTalk
• Support Desk, AI-assisted tickets and quotes
• Projects & Tasks, Kanban boards where agents execute tasks automatically
• Wiki, team knowledge base agents can read and update
• Scheduled Prompts, run agents on a schedule
• Integrations, n8n, GitHub, MCP servers, S3-compatible storage
• Skills & Extensions, reusable capability modules and a marketplace
• Guardrails and approvals, rules for what an agent may see or say, and a human yes before risky actions
• Runs, a step-by-step record of everything an agent did
• Governance, teams, RBAC, audit log, and usage analytics
• API & Webhooks, automate the platform itself

Create and manage AI agents with:

• Custom system prompts
• Model selection per agent
• Tool access control (allowlist/denylist)
• Skills attachment
• Subagent hierarchies
• MCP server connections
• YAML-based configuration

• Install a ready-made agent (support, sales, finance, ops) into your workspace in one click
• Or create an agent from scratch and shape it yourself
• Customize the model, system prompt, tools, and skills on any agent, prebuilt or your own
• Add subagents and delegation to the ones you install
• New agents keep arriving as the library grows

Chat with your agents directly in the browser:

• Persistent sessions with full history
• File attachments and rich responses
• Per-session model and agent switching
• Command palette (Cmd+K) for quick navigation across the whole studio

Supported providers:

• OpenAI, Anthropic, Gemini, DeepSeek, NVIDIA, via API key
• Gemini and Qwen, via OAuth login
• OpenRouter, one key, many models
• llama.cpp server, self-hosted local models
• Custom LLM, any OpenAI-compatible endpoint

• Put an approval gate on risky actions: payments, sends, deletes, publishing
• The agent pauses, asks for a decision, and only proceeds once you approve
• Approval requests are delivered to Slack or Telegram
• Set an approval destination per agent, and per subagent
• Denied actions stop the run cleanly, nothing happens without a yes

Five rule types:

• Keyword, blocks text containing any term on a list. Matching ignores case unless you turn that off
• Pattern, blocks text matching a regular expression
• Personal info, blocks text that looks like an email address, a phone number, or a social security number
• Category, a keyword list with a label, for example "medical_advice". The label shows in the block reason and does not change matching
• Written policy, you describe the rule in plain language, up to 2,000 characters, and Oido's built-in judge decides whether a message breaks it

Direction decides which side of the conversation a rule checks:

• Input, what people send to the agent
• Output, what the agent is about to send back
• Input and output, both

Scope decides who a rule applies to:

• Organization-wide, every agent in your workspace
• Agent-specific, only the agents you pick. An agent-specific rule can add to the organization's rules but never loosen them

How checks behave:

• Keyword, pattern, personal info, and category rules always run
• Written-policy rules call the judge, so they only run when you have configured at least one for that agent and direction
• If a rule cannot be evaluated because of an infrastructure error, the message is allowed through rather than blocking everyone. Use keyword, pattern, or personal info rules for anything that must never get through
• Blocks are recorded in the audit log

Open Runs in the sidebar to see them all, and filter by outcome. Open one run to see its trajectory, every step the agent went through, in order:

• The system prompt and the message that started it
• Each model response, with its usage and timing
• Each tool call, with what was sent, what came back, and how long it took
• Each approval request and the answer it got
• Each subagent it handed work to
• Any point where the conversation was compacted to save context

Multi-user support with role-based access control:

• Roles with granular permissions
• Team management
• User invitations
• Session isolation per user

Full visibility into what agents and users do:

• Audit log of every significant action across the platform, including messages blocked by a guardrail
• Usage analytics: tokens, turns, and cost per agent and per model
• Filter by user, agent, and time range

Connect agents to messaging platforms:

• Slack
• WhatsApp
• Telegram
• Discord
• WeChat
• DingTalk

Channels support:

• Sender policies (whitelist/any)
• Session scoping (user/global)
• Dispatch modes (sequential/parallel)
• Per-channel model and instruction overrides
• Pairing flow for device authorization

Built-in AI-assisted customer support:

• Ticket management with statuses and assignment
• Quote creation and tracking
• Agents draft responses and triage incoming requests
• Full conversation history per ticket

Team knowledge base built into the platform:

• Markdown pages with slugs and linking
• Agents can read wiki content to answer questions
• Agents can create and update pages, keeping docs current automatically

Run agents on a schedule:

• Cron-style recurring prompts
• Assign any agent to a schedule
• Results delivered to sessions or channels

Built-in Kanban board for project management:

• Custom columns per project
• Task creation with descriptions
• Agent assignment per task
• Auto-run on task move (agents execute tasks automatically)
• Reorderable columns and tasks

Native Model Context Protocol support for extending agent capabilities:

• stdio transport (local processes)
• SSE transport (remote servers)
• Per-server environment variables
• OAuth configuration
• User/project scoping

Skills are reusable capability modules attached to agents:

• Built-in skills (DuckDuckGo search, Gmail, Google Calendar, n8n, web fetch)
• User-installed skills from SKILL.md files
• Extension marketplace support

Extensions add platform-level functionality:

• Install from git URLs, npm packages, local paths
• Configure via settings UI
• Enable/disable per deployment

Deep integration with n8n workflow automation:

• Connect n8n instance via API
• Browse and trigger workflows
• View execution history
• Manage credentials
• Agents can create, update, list, and delete workflows
• Data table operations (create, read, upsert, delete rows)

• Oido Trigger: start a workflow on an agent dispatch or an Oido event (agent run finished, channel message, ticket created)
• Oido: OCR a document, read or ingest a wiki page, embed, rerank, and respond back to the agent that dispatched the run
• Authenticates with an Oido API key

Connect repositories to agents:

• Authenticate with GitHub PAT
• Register repositories for agent access
• Assign repos to specific agents
• Agents can read code, manage issues, and review PRs

S3-compatible storage via RustFS:

• File upload and download
• Bucket management
• Object listing
• Configurable storage backend

Automate the platform itself:

• REST API (v1) with API key authentication
• Webhooks for reacting to platform events
• Hooks for customizing agent behavior

• Hosted and maintained by Oido Systems
• Enterprise-grade security and encryption
• 99.9% uptime SLA
• Automatic updates and backups
• SOC 2 compliance (in progress)