BrightBean Studio: Open-source social media management platform built with AI agents

What BrightBean Studio does
BrightBean Studio is an open-source, self-hostable social media management platform built for creators, agencies and SMBs. It provides a free alternative to commercial tools like Buffer, Sendible, and SocialPilot without per-seat, per-channel, or per-workspace limits.
Platform support
The platform supports publishing, comments, DMs, and insights across these platforms:
- Facebook (all features)
- Instagram (all features)
- Instagram Personal (all features)
- LinkedIn Personal (publish, comments, insights)
- LinkedIn Company (publish, comments, insights)
- TikTok (publish, comments, insights)
- YouTube (publish, comments, insights)
- Pinterest (publish, insights)
- Threads (publish, comments)
- Bluesky (publish, comments)
- Google Business Profile (publish, insights)
- Mastodon (publish, comments)
Key features
- Multi-workspace & teams: Unlimited orgs → workspaces → members with granular RBAC, custom roles, invitations, and separate Client role for external collaborators
- Content composer: Rich editor with per-platform caption/media overrides, version history, reusable templates, content categories & tags, and Kanban idea board
- Calendar & scheduling: Visual calendar with recurring weekly posting slots per account and named queues that auto-assign posts to the next available slot
- Publishing engine: Direct first-party API integrations (no aggregator), automatic retries, per-account rate-limit tracking, and 90-day publish audit log
- Approval workflows: Configurable stages (none/optional/internal/internal + client), threaded internal & external comments, reminders, and full audit trail
- Unified social inbox: Comments, mentions, DMs, and reviews from every connected platform in one place with sentiment analysis, assignments, threaded replies, and historical backfill
- Media library: Org- and workspace-scoped libraries with nested folders, auto-generated platform-optimized variants, and alt text
- Client portal: Passwordless 30-day magic-link access for clients to approve or reject posts without creating an account
- Notifications: In-app, email, and webhook delivery with per-user preferences for every event type
- Security & ops: Encrypted token & credential storage, optional 2FA (TOTP), Google/GitHub SSO, Sentry support, and 7-day reversible org-deletion grace period
- White-label friendly: Per-workspace branding (logo, colors) and workspace defaults for hashtags, first comments, and posting templates
Deployment options
The platform can be deployed with one-click buttons on Heroku, Render, or Railway, run on your own VPS via Docker, or run locally. All platform integrations use your own developer credentials to talk directly to official first-party APIs, avoiding aggregator middlemen and vendor lock-in.
Development context
This tool was built in 3 weeks using Claude and Codex, demonstrating how AI coding agents can accelerate development of complex applications. The project structure includes typical Python/Django components with Docker support and pre-commit hooks.
📖 Read the full source: HN AI Agents
👀 See Also

Benchmark shows AI browser automation tools vary 2.6x in token costs despite identical accuracy
A benchmark of 4 CLI browser automation tools using Claude Sonnet 4.6 on 6 real-world tasks found all achieved 100% accuracy, but openbrowser-ai used 36,010 tokens while others used 77,123-94,130 tokens. Tool call count was the strongest predictor of token cost.

LUMA SOUL: Claude-Powered Minds with Permanent Creator Lock and Transparent Memory
LUMA SOUL is a presence platform where Claude minds get portraits, voices, and soul documents. Key design: creators lose edit rights permanently at submission, and memory is fully transparent.

TOON MCP server reduces tool result tokens by 30-60% in OpenClaw
An MCP server that compresses structured JSON tool results into the TOON format can cut token usage by 30-60% for tabular data like database queries and API responses, helping delay context window compaction in OpenClaw sessions.

Scaling Karpathy's Autoresearch with 16 GPUs: Results and Methods
The SkyPilot team gave Claude Code access to 16 GPUs on a Kubernetes cluster to run Karpathy's Autoresearch project. Over 8 hours, the agent submitted ~910 experiments, reduced validation bits per byte from 1.003 to 0.974 (2.87% improvement), and reached the best validation loss 9x faster than sequential execution.