command
Bundle OKF 0.1 · 10 conceitos · cinatra-ai/cinatra
Open source Repository Open in the app JSON README (API)
About
# command
* [Author a concept with the memory command](/command/author-a-concept-with-the-memory-command.md) - Write concepts through the `memory add` command rather than by hand, because it enforces the identity, containment, and size rules for you.
* [Check the bundle after every write](/command/check-the-bundle-after-every-write.md) - Run `memory check` after you add or edit a concept.
# convention
* [Check the index for a duplicate before you write](/convention/check-the-index-for-a-duplicate-before-you-write.md) - Read `index.md`, or run `memory recall` on the words of your intended title, before you add a concept.
* [Choose the type field deliberately](/convention/choose-the-type-field-deliberately.md) - Set frontmatter `type` to the kind that describes the insight, from `Convention`, `Correction`, `Command`, and `Debugging Insight`.
* [Never write a secret into a concept](/convention/never-write-a-secret-into-a-concept.md) - An API key, a token, a password, a private URL with
Details
- Kind
- OKF bundles
- Topic
- AI, RAG & memory
- Publisher
- cinatra-ai
- Origin
- okf_github
- Category
- dados
- Version
- 0.1
- Stars
- 7
- Forks
- 1
- Open pull requests
- 36
- Last push
- 2026-09-10T05:45:47Z
- Repository state
- ativo
- Language
- TypeScript
- License
- Apache-2.0
- Added
- 2026-09-09 05:03:59
- Updated
- 2026-09-09 05:03:59
- Origin id
cinatra-ai/cinatra:packages/memory/seed/conventions/index.md
README
<div align="center"> <img src="src/app/icon.svg" width="96" alt="Cinatra logo" /> # CINATRA **The open source AI workspace for teams** [](LICENSE) [](https://docs.cinatra.ai) [](https://github.com/cinatra-ai/cinatra/actions/workflows/build-image.yml) [](https://docs.cinatra.ai) <img src="screenshot.png" alt="Cinatra — the open source AI workspace for teams" width="820" /> </div> > [!WARNING] > **Cinatra is not production ready!** > It is under active development and has not been hardened, security-audited, or stability-tested for production workloads. Run it for evaluation, local development, and self-hosted experimentation only. Do not deploy it to handle production data, untrusted users, or business-critical workflows yet. APIs, schemas, and the extension contract may change without notice. --- ## What Cinatra is Cinatra is an open source AI workspace for teams: a shared, persistent, browser-based environment where people, AI assistants, and autonomous agents work together. It turns isolated prompts into durable workflows that have state, tooling, handoffs, approvals, and real operational outputs. Most AI tools are optimized for individual use and short chat sessions. Cinatra is built for collaborative work that takes time, spans systems, requires oversight, and improves as teams capture their patterns into reusable skills, extensions, artifacts and agents. ### Core pillars - **[A built-in AI assistant](https://docs.cinatra.ai/guides/user/built-in-ai-assistant/).** The workspace has a chat assistant that works across its MCP-exposed capabilities — agents, connectors, data, dashboards, files, and artifacts — scoped to what you can access. Describe what you need to build or run, and it can draft agents, workflows, dashboards, and artifacts, dispatch agents, and coordinate with other @-mentioned assistants in the same thread. - **[A connected ecosystem of capabilities](https://docs.cinatra.ai/guides/user/connected-ecosystem/).** Agents, connectors, skills, objects, lists, and dashboards live in one capability fabric. Workflows compose across domains instead of living as isolated scripts. - **[Human-in-the-loop by design](https://docs.cinatra.ai/guides/user/human-in-the-loop/).** Agents pause at typed HITL gates for review, edits, approvals, or missing context — instead of forcing an all-or-nothing automation model. - **[Continuous learning](https://docs.cinatra.ai/guides/user/continuous-learning/).** Prompt edits inside HITL surfaces are captured into per-user, per-agent personal skills that prime the next run. Improvements become reusable operating knowledge instead of getting lost in chat history. - **[An extendable marketplace](https://docs.cinatra.ai/guides/user/marketplace-and-extensions/).** Install agents, connectors, skills, artifacts, and workflows from the marketplace onto a running workspace, with the access each team grants — and publish your own. - **[Cross-instance collaboration](https://docs.cinatra.ai/guides/user/cross-instance-collaboration/).** Cinatra instances share a marketplace, install each other's extensions, and call each other's agents over A2A — with run data staying where the run runs. - **[Durable workflows](https://docs.cinatra.ai/guides/user/durable-workflows/).** Background execution with BullMQ over Redis and durable state in PostgreSQL. Workflows survive page reloads, resume from network drops, and pause for human approval without losing context. ## Example: Email outreach campaign 1. A user tells the AI assistant they need to run an outreach campaign. 2. Cinatra spins up an Email Outreach agent (or reuses an existing one) with a default `SKILL.md` describing the ideal customer profile and contact-selection rules. 3. The agent selects prospects, enriches their data, and drafts personalized emails using company context and recent events. 4. The team reviews the drafts, edits them one by one, or applies prompt-driven changes across the whole batch. 5. Replies are tracked and follow-ups are sent automatically using the same workflow. 6. Prompts and edits made along the way can be captured back into a custom `SKILL.md`, so the next campaign starts tuned to the team's real working style. ## Inside the app The main sidebar groups the day-to-day workspace: - **Intelligence → Chat** — multi-threaded AI assistant chat with team threads; the place agents are created, run, and edited conversationally - **Agents** — two tabs: the top-level `/agents` route is **All Agents**, the run-agent picker; **Executions** (`/agents/executions`) is an interactive dashboard of recently used and recently run agents. Installing agents happens in Configuration → Marketplace - **Management** — Personal, Projects, Teams, Organizations - **Information** — Artifacts, Data (a unified object list with typed views, plus History and Merge), and Analytics (LLM and API usage) - **Tools** — Skills (catalog, installed packages, match overview, autosave from chat edits) and Connectors (e.g. Gmail, Google Calendar, Apollo, LinkedIn, WordPress, Drupal, Apify, YouTube, GitHub) Platform admins also see an **Admin** group at the top of the sidebar (Approvals, Configuration). Beyond the sidebar, the platform ships routes for **Dashboards** (operator workspaces composed from extension-shipped portlets, with the agents dashboard at `/agents/executions`, also reachable as an MCP primitive) and **Notifications** (a durable feed with real-time updates and failure routing). They are reachable directly by URL. A separate **Configuration** area (linked from the sidebar as **Admin → Configuration**) covers platform-level settings: environment, AI providers, MCP, extensions, webhooks, [marketplace](https://docs.cinatra.ai/guides/admin/marketplace/) (install agents, connectors, skills, artifacts, and workflows from the shared registry), skills, [permissions](https://docs.cinatra.ai/guides/admin/permissions/) (a co-owner model across extension resources — agents, agent runs, connectors, skills, skill packages, artifacts, and workflows), access control, workflows, agents, assistants, workspace, telemetry, and development. Most configuration screens are admin-only. The [Admin Guide](https://docs.cinatra.ai/guides/admin/) covers this surface in detail. --- ## Architecture Cinatra is a monorepo of TypeScript packages running on Next.js. Each package owns its persistence, background jobs, React screens, and capability surface. Packages communicate through public capability surfaces rather than importing each other's internals. Cinatra speaks four open agent protocols so that agents authored here are not locked here, and so that agents from other platforms plug in without bespoke integration: - **OAS (Open Agent Specification / agentspec)** — every agent is a declarative OAS Flow file - **A2A (Agent-to-Agent)** — every agent is callable from any A2A client; Cinatra calls remote A2A agents as local tools - **AG-UI (Agent-User Interaction Protocol)** — typed lifecycle events streamed over SSE with durable replay - **A2UI (Agent-to-User Interface)** — declarative HITL surface payloads on a parallel channel Every capability is also an MCP primitive, and the whole workspace is itself reachable as an OAuth-secured MCP server — so external MCP clients (Claude Desktop, ChatGPT, OpenAI Codex, Claude.ai) can drive Cinatra directly. The agent runtime is **WayFlow**, the reference OAS implementation, running as a Python sidecar. The Next.js app invokes it over A2A — making the runtime replaceable by any OAS-compliant alternative. For the full write-up, see the [Architecture](https://docs.cinatra.ai/references/platform/architecture/), [Open standards in Cinatra](https://docs.cinatra.ai/references/platform/open-standards/), and [MCP reference](https://docs.cinatra.ai/references/mcp/) pages. --- ## Quick start ```bash npx @cinatra-ai/cinatra install ``` The [cinatra CLI](https://www.npmjs.com/package/@cinatra-ai/cinatra) is the single, idempotent command that takes a machine from zero to a running instance: it checks prerequisites (Node.js 24+, git, pnpm via Corepack, Docker + Compose), clones Cinatra, creates your `.env.local`, brings up the local Docker services, installs dependencies, and runs first-time setup. Re-running it on an existing checkout reconciles it in place instead of cloning again. Use `--mode prod` for a production instance. Once it finishes, `cd` into the checkout (`cinatra/` by default) and run `npx @cinatra-ai/cinatra instance start` for day-to-day dev-server start/stop, or `npx @cinatra-ai/cinatra doctor` / `npx @cinatra-ai/cinatra status` to check on it. After pulling new code, `npx @cinatra-ai/cinatra instance refresh` reconciles dependencies and the dev database schema to match your checkout; `npx @cinatra-ai/cinatra update` moves the checkout itself forward first (dev → latest `main`, prod → latest release) and then reconciles. Install the CLI globally with `npm install -g @cinatra-ai/cinatra` to drop the `npx` prefix. Open <http://localhost:3000>. The first user to register becomes the platform admin and lands in the in-app setup wizard for the remaining first-run configuration. Full walkthrough with prerequisites, services, and first-time configuration: see [Installation](https://docs.cinatra.ai/guides/hosting/installation/) and [Quickstart](https://docs.cinatra.ai/guides/hosting/quickstart/) in the Hosting Guide. --- ## Documentation The full documentation set is published at **[docs.cinatra.ai](https://docs.cinatra.ai)**. Release history and notable changes are tracked in **[CHANGELOG.md](CHANGELOG.md)**; each tagged release also has auto-generated notes on the [GitHub Releases](https://github.com/cinatra-ai/cinatra/releases) page. --- ## Contributing Issues and pull requests are welcome — start with **[CONTRIBUTING.md](CONTRIBUTING.md)** and the **[Code of Conduct](CODE_OF_CONDUCT.md)**. To report a security vulnerability, see **[SECURITY.md](SECURITY.md)**. ## License Cinatra is open source under the Apache License 2.0 — see **[LICENSE](LICENSE)**.