OpenIntel
CrackedNut/OpenIntel
Agentes de código autônomos no seu chat com dashboard de controle
Ver no GitHubPlataforma que transforma Claude Code em agentes autônomos chat-nativos. Executa em Mattermost, Slack ou Discord com dashboard local para gerenciar persona, skills, projetos e logs em tempo real, permitindo aprovação de ações via emoji e paralelização de sessões.
Instalação
npx convez-ai CrackedNut/OpenIntel1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Orquestração de tarefas de desenvolvimento em equipe via chat
- Automação de operações DevOps com aprovação manual em tempo real
- Execução de análises de código e refatoração assistida por IA em threads paralelas
Veredito

Solução robusta para equipes que desejam agentes de código conversacionais com governança. O setup one-liner e dashboard integrado reduzem fricção operacional, mas exige conforto com TypeScript, CLI e gerenciamento de infraestrutura local.
README
A tradução para o português deste README ainda está na fila.
OpenIntel
✴ ▄▀▄ ▀█▀ ✴ OpenIntel
✴ █ █ █ ✴ autonomous Claude Code agents in your chat,
✴ ▀▄▀ ▄█▄ ✴ with a dashboard to run the whole operation
OpenIntel turns Claude Code into a team of chat-native agents. Run the bot on any machine, point it at a Mattermost, Slack, or Discord channel, and the whole team can talk to an autonomous coding agent — watch it work in real time, approve its actions with emoji, spin up parallel sessions in threads, and manage everything (persona, skills, projects, config, logs) from a local web dashboard.
Built on a fork of anneschuth/claude-threads (Apache-2.0), extended into a self-contained agent platform.
Install (one-liner)
curl -fsSL https://raw.githubusercontent.com/CrackedNut/OpenIntel/main/install.sh | bash
On a fresh machine this installs bun + the Claude Code CLI if missing, clones and builds OpenIntel, runs an interactive wizard for your bot token/channel, seeds an editable agent persona, starts the daemon, and prints the dashboard URL. Each device gets its own agent for its own project.
Already installed? Manage everything with the openintel command (alias: claude-threads):
openintel status # daemon, branch/commit, version
openintel panel # open the dashboard
openintel logs # tail the bot log
openintel setup # reconfigure (tokens, channels, users)
openintel install # pull latest, rebuild, restart
openintel rollback # restore the previous build
openintel start|stop|restart
The Dashboard
A local web panel at http://127.0.0.1:7777 (binds localhost only), served by the bot itself:
- Overview — live sessions with model / context / cost, working-vs-idle state, per-session Interrupt / Stop controls, platform connection status
- Persona — edit the agent's
SOUL.md(identity/tone) andDIRECTIVES.md(hard guardrails); new sessions pick changes up automatically - Projects — manage the projects index injected into the agent's system prompt
- Skills — manage the agent's skill library (
SKILL.mdplaybooks, flat orcategory/skilllayouts) - Config — edit the raw
config.yamlwith validation - Paths — point persona/projects/skills at any directory per machine
- Logs — live tail of the daemon log with filtering
- Update & restart — one click pulls the latest main, rebuilds (with automatic snapshot for rollback), and restarts; sessions persist and resume
In Chat
Mention the bot at the channel root and it becomes a shared channel agent — everyone allowed can talk to it, no thread required:
@yourbot what's the state of the deploy?
Spin up parallel thread sessions (each with its own Claude instance) without leaving the channel:
!thread fix the auth bug # fresh thread session on a topic
!thread fix the auth bug -history # seeded with recent channel conversation
The bot 👀-reacts when it accepts your message and flips it to ✅ when the turn completes, shows typing indicators, and streams output live.
Core features
- Real-time streaming of Claude's responses, tool use, diffs, and task lists into chat
- Multi-platform — Mattermost, Slack, and Discord, multiple workspaces/servers simultaneously
- Per-session model picker —
!modellists models to react-pick for this session only;!model --defaultalso sets the default for new sessions - Concurrent sessions — channel-mode + any number of thread sessions, persisted across restarts
- Permission modes —
default(every action prompts 👍/✅/👎),auto(classifier auto-approves low-risk),bypass; switch in-session with!permissions <mode> - Collaboration —
!invite @user/!kick @user; collaborators land asCo-Authored-By:trailers on commits - Git worktrees —
!worktree feature/fooisolates the agent's changes (alsolist,switch,remove,cleanup,off) - Files both ways — drop any file into chat (100 MB cap) for Claude to read; Claude posts screenshots/PDFs/plots back via
send_file - Permalink reading — paste a Mattermost/Slack/Discord permalink and Claude resolves it via
read_post - Multi-account Claude (opt-in) — round-robin sessions across subscriptions/API keys with rate-limit cooldown
Session Commands
Type !help in any session:
| Command | Description |
|---|---|
!thread <topic> [-history] | Spawn a new thread session from channel mode (own Claude instance) |
!help | Show available commands |
!context / !cost / !compact | Context usage / token cost / compress context |
!cd <path> | Change working directory (restarts Claude) |
!permissions <mode> | Set permission mode: default / auto / bypass |
!model [--default] | React-pick a model for this session (--default also saves it as the default) |
!worktree <branch> | Git worktree management (also: list, switch, remove, cleanup, off) |
!plugin <list|install|uninstall> [name] | Manage Claude Code plugins (restarts Claude) |
!invite @user / !kick @user | Session collaboration |
!github-email <email> | Register your GitHub noreply email for commit attribution |
!queue <msg> / !steer <msg> | Buffer a message for when Claude is free / interrupt and redirect |
!search <query> | Search the bot's session archives |
!bug <desc> | Report a bug with context |
!approve | Approve pending plan (alternative to 👍) |
!escape | Interrupt current task (session stays active) |
!stop | Stop this session |
!kill | Emergency shutdown (ALL sessions + bot) |
Interactive Controls
- Permission approval: 👍 allow once · ✅ allow all · 👎 deny
- Plan approval: 👍 approve · 👎 request changes
- Questions: 1️⃣ 2️⃣ 3️⃣ 4️⃣ for multiple choice
- Session control: ⏸️ interrupt · ❌/🛑 stop · 🔄 resume a timed-out session
Agent Identity (Persona / Skills / Projects)
Every session's system prompt is assembled from editable content, resolved per machine:
| Content | What it is | Default location |
|---|---|---|
SOUL.md | Who the agent is — name, voice, tone | ~/.config/claude-threads/agent/SOUL.md |
DIRECTIVES.md | Hard behavioral rules | ~/.config/claude-threads/agent/DIRECTIVES.md |
| Projects index | One description.md per ongoing project | ~/.config/claude-threads/agent/projects/ |
| Skills | SKILL.md playbooks the agent can invoke | ~/.config/claude-threads/agent/skills/ |
Legacy locations (~/.hermes/*, ~/agent-memory/projects, ~/.claude/skills) are auto-detected if present. Override any path in the dashboard's Paths tab.
Prerequisites
- Bun 1.2.21+ (the installer handles this) — or Node 20+
- Claude Code CLI logged in — test with
claude --version(the installer installs it; runclaudeonce to authenticate) - A bot on your platform — see SETUP_GUIDE.md. For Discord, the bot must have the MESSAGE CONTENT privileged intent enabled (Developer Portal → Bot → Privileged Gateway Intents) or it can't read messages.
Architecture (short version)
One daemon process per machine. Each chat session spawns its own claude CLI process with a per-session MCP server that routes permission prompts, file uploads, and permalink reads back through the platform. Sessions are keyed platform:thread (or platform:channel for the shared channel agent) and persist across restarts. The dashboard runs inside the bot process. See CLAUDE.md for the full architecture and development guide, and SETUP_GUIDE.md for platform bot creation.
Development
bun install
bun run build # bundle to dist/
bun test # ~2,600 unit tests
bun run lint
Deployment on a managed machine is via openintel install [ref] — it snapshots the current build first, so openintel rollback always works.
Config and data live under
~/.config/claude-threads/and~/.claude-threads/— the legacy directory names are kept so existing deployments survive upgrades.
Credits & License
OpenIntel is a hard fork of claude-threads by Anne Schuth — the streaming engine, platform layer, session lifecycle, and permission system started there. Licensed Apache-2.0, same as upstream.