hive
Multi-Agent Harness for Production AI
- 11.1k estrellas
- 5.6k forks
- 1093 issues abiertos
- Lenguaje: Python
- Licencia: Apache-2.0
- actualizado
Instalación
git clone https://github.com/aden-hive/hive && cp hive/*.md ~/.claude/agents/- Clone the repository and copy the agent .md definitions into ~/.claude/agents (or .claude/agents inside a project).
- Start a new Claude Code session to load the agents.
- Delegate work to them with the Task/Agent tool or by name.
Nota editorial
Hive is a Python-based multi-agent execution harness designed to run complex, long-horizon business workflows in production environments. Rather than requiring users to write orchestration code, it accepts a plain objective and compiles a graph-based directed acyclic graph (DAG) that coordinates specialized subagents running tasks in parallel. It connects to Claude through the Anthropic API alongside OpenAI and Google Gemini, and exposes 102 MCP tools for integrations. Key runtime features include role-based persistent memory that evolves with a project, deterministic fault tolerance with crash recovery, session isolation, shared buffers between agents, cost enforcement, and audit trails. A browser-use capability allows agents to interact with live web interfaces. The project also ships HoneyComb, a community dashboard that tracks which job categories AI agents are automating, framed as a token-based prediction market. Hive targets engineering teams that have moved past single-agent prototypes and need uptime, observability, and human-in-the-loop controls at scale.
24 items en este repositorio
SOP for debugging browser automation failures on complex websites. Use when browser tools fail on specific sites like LinkedIn, Twitter/X, SPAs, or sites with Shadow DOM.
Required before any hive-browser CLI command. The browser is driven from the terminal by running `hive-browser <command> ... --json` via terminal_exec — not via MCP tools. Teaches the browser lifecycle rules (the bridge attaches to the USER'S running Chrome — never kill or launch browser processes; timeouts are transport issues, not crashes), the screenshot + coordinate workflow (hive-browser interact with a fractional coordinate) that reaches shadow-DOM inputs selectors can't see, the viewport-fraction coordinate rule (not pixels), rich-text editor quirks ("send button stays disabled" failures), and CSP gotchas. Covers Chrome via CDP through the GCU Beeline extension.
Required reading whenever any chart_* tool is available. Teaches the one-tool embedding contract (call chart_render → live chart appears in chat AND a downloadable PNG lands in the queen session dir), the ECharts (data viz) vs Mermaid (structural diagrams) decision, the BI/financial-grade aesthetic baseline (no chartjunk, restrained palette, proper typography, single message per chart), and the canonical spec patterns for the 12 most-common chart types. Skipping this leads to 1990s-Excel charts, missing downloads, and the agent writing markdown image links by hand instead of letting chart_render drive the UI.
Claim tasks, record step progress, and verify SOP gates in the colony SQLite queue. Applies when your spawn message includes a db_path field.
Proactively extract critical values from tool results into working notes before automatic context pruning destroys them.
Follow a structured recovery decision tree when tool calls fail instead of blindly retrying or giving up.
Required before calling image_generate. Create and edit images from a prompt — generate an image, make a picture / logo / illustration / icon / banner / poster / thumbnail / hero image / mockup / product shot / social graphic, or edit / restyle / combine existing images from reference images. Uses OpenAI gpt-image-2 through the Hive image service, billed to the user's Hive credits like an LLM call (no API key needed). Teaches the exact call shape, the quality/cost tradeoff (quality="low" is the default and cheapest), reference-image editing, how to show the result to the user with attach_file, and the failure modes (out of credits, model unavailable, moderation).
Read before automating LinkedIn with browser_* tools. LinkedIn combines shadow DOM (#interop-outlet), strict Trusted Types CSP that silently drops innerHTML, Lexical composer, native beforeunload dialogs that hang the bridge, and aggressive spam filters — each has bitten us at least once. Verified flows for profile messaging, connection-request acceptance, feed composition, and search. Requires hive.browser-automation. Verified against logged-in production 2026-04-11.
Maintain a free-form scratchpad of decisions, extracted values, and open questions so context pruning doesn't lose anything you still need.
Read, write, merge, split, rotate, watermark, encrypt, and OCR PDF files using Python (pypdf, pdfplumber, reportlab, pypdfium2) and command-line tools (poppler-utils, qpdf). Use when the user asks to extract text/tables/images from a PDF, create or modify a PDF, combine or split PDFs, OCR a scanned PDF, password-protect a PDF, or render PDF pages as images.
Periodically self-assess output quality to catch degradation before the judge does.
Set up a Slack notification channel (Sentinel) for a colony by driving the browser — reuse or create the "Hive Sentinel" Slack app from a JSON manifest, install it, capture the bot + app tokens, create/select the channel via the Slack API, and turn Sentinel on so the colony can ping the user on Slack and accept replies. Use when the user asks to "set up Slack notifications", "get pinged on Slack", "connect Slack for alerts", "set up Sentinel on Slack", or clicks the in-app "Set this up with the agent" button on the Slack channel step. Requires hive.browser-automation.
Set up a Telegram notification channel (Sentinel) for a colony by driving the browser — create a bot via @BotFather in Telegram Web, store its token, detect the chat to notify, and turn Sentinel on so the colony can ping the user on Telegram and accept replies. Use when the user asks to "set up Telegram notifications", "get pinged on Telegram", "connect Telegram for alerts", "set up Sentinel on Telegram", or clicks the in-app "Set this up with the agent" button on the Telegram step. Requires hive.browser-automation.
Required reading whenever any shell_* tool is available. Teaches the foreground/background dichotomy (terminal_exec auto-promotes past 30s, returns a job_id you poll with terminal_job_logs), the standard envelope shape (exit_code, stdout, stdout_truncated_bytes, output_handle, semantic_status, warning, auto_backgrounded, job_id), output handle pagination via terminal_output_get, when to read semantic_status instead of raw exit_code (grep/rg/find/diff/test exit 1 is NOT an error), the destructive-warning surface (rm -rf, git push --force, DROP TABLE), tool preference (the terminal handles ALL file read/write/edit/search; use gcu-tools / hive_tools where they fit), and the bash-only-on-macOS policy. Skipping this leads to "tool returned no output" surprises, orphaned jobs, and panic over benign grep exit codes.
Use terminal_rg / terminal_glob for all filesystem search — your project tree as well as system configs, /var/log, /etc, archive contents. Teaches the rg vs glob vs terminal_exec("find/ls/du/tree") split, common rg flag combos for code/logs/configs, glob patterns for finding files by name, the rule that mtime/size/type predicate queries drop to terminal_exec("find ..."), and that for tree views or single-file stat info you should just use terminal_exec instead of inventing a tool. Read before reaching for raw shell to grep or find anything.
Use when launching anything that runs longer than a minute, anything that streams logs, anything you want to keep running while doing other work — or when terminal_exec auto-backgrounded on you and returned a job_id. Teaches the start→poll→wait pattern with terminal_job_logs offset bookkeeping, the `wait_until_exit=True` blocking-poll idiom, the truncated_bytes_dropped resumption signal, the merge_stderr decision, the SIGINT→SIGTERM→SIGKILL escalation ladder via terminal_job_manage, and the hard rule that jobs die when the terminal-tools server restarts. Read before calling terminal_job_start, or right after terminal_exec auto-backgrounded.
Use when you need state across calls — building env vars, navigating with cd, driving REPLs (python -i, mysql, psql, node), or responding to interactive prompts (sudo password, ssh host-key confirmation, mysql connection). Teaches the prompt-sentinel exec pattern (default mode), raw I/O for REPLs (raw_send=True then read_only=True), the one-in-flight-per-session rule, and the close-or-leak-against-the-cap discipline. Bash on macOS — never zsh; explicit shell=/bin/zsh is rejected. Read before calling terminal_pty_open.
Read when a terminal-tools call returned something surprising — empty stdout despite no error, exit_code is null, output_handle came back expired, "too many jobs" / "session busy" / "too many PTYs", warning was set unexpectedly, semantic_status disagrees with exit_code. Diagnostic recipes only — load on demand. Don't preload; the foundational skill covers the happy path.
Concrete patterns for breaking colony work into parallel worker jobs via run_playbook — when fan-out helps, how to model the goal as a tracker table, write the worker skill, author the playbook, pilot, and let convergence retry/resume the gap.
Author a new Agent Skill for a Hive agent that conforms to the Agent Skills specification (SKILL.md with YAML frontmatter, optional scripts/references/assets directories). Use when the user asks to create, scaffold, add, or package a new skill for a Hive agent.
Read before automating X / Twitter with browser_* tools. Verified flows for post, reply, delete, search-and-engage, plus the Draft.js compose quirks that silently disable the send button. Includes the daily-reply and job-market-reply playbooks. Requires hive.browser-automation for the underlying screenshot + coordinate workflow. Verified 2026-04-11.
Resumen de Subagents
English | 简体中文 | Español | हिन्दी | Português | 日本語 | Русский | 한국어
An open-source runtime for colonies of AI agents that do real business work.
Tell a queen agent what you need. She does the first piece herself to learn what the job takes, then hands the rest to as many parallel workers as it needs. Every result lands in a shared tracker you can query, audit and resume from.
A real run. Only the typing and the waiting are sped up, and the badge says by how much. Watch the full run (MP4).
What you just watched
A growth team preparing a Show HN wants to know how five developer tools' own launches did on Hacker News. Here's what Hive did with that one message:
You hand it to a queen. Queens are long-lived agents with a role (here, Head of Growth) and their own memory and tools.
She asks the one question that changes the answer: every launch post, or only each product's original launch? You pick every launch, and she carries on.
She proposes a colony, and you approve it. Five products means five parallel jobs, so the chat becomes a colony: the queen plus as many worker agents as the work needs.
She does one herself. She takes PostHog first, works out what counts as a genuine launch post rather than a passing mention, checks the numbers against the Hacker News API, and saves the method as a reusable skill.
She fans out. A playbook starts one worker per product. Each follows her skill and writes its result to the colony's tracker, a shared SQLite table.
She checks every row before she answers, and fixes one. Cal.com's worker counted a post from its old name, Calendso, but that post wasn't a Show HN, Launch HN or YC launch. She clears the row, and the chart shows Cal.com as not found rather than zero.
None of this was wired up in advance. There's no workflow graph to draw: the queen shapes the colony as the work unfolds, and a tracker on disk, not a context window, keeps count of what's done and what's left.
Quick start
You'll need Python 3.11+, Node.js 20+ and git. The quickstart installs uv and ripgrep if they're missing, and offers to install Node.
And a model. The quickstart sets up any of these:
an API key from Anthropic, OpenAI, Google Gemini, Groq, Cerebras or OpenRouter
a coding subscription you already pay for: Claude Code, OpenAI Codex, Kimi Code, MiniMax, Z.AI or Antigravity
Hive LLM
a local model through Ollama, with no key at all
git clone https://github.com/aden-hive/hive.git
cd hive
./quickstart.sh # macOS / Linux
.\quickstart.ps1 # Windows (PowerShell 5.1+)The quickstart creates a single Python environment for the workspace, stores your API key encrypted under ~/.hive, asks which model to use, builds the dashboard and opens it at http://127.0.0.1:8787. Next time, run hive open from the repository.
Note
This repository is a
uvworkspace:pip install -e .at its root installs a placeholder that won't run, so use the quickstart. Without the repository,pip install "aden-hive[tools]"installs Hive from PyPI; then runhive open. To add an optional package (a chat-app channel, for example), see Optional packages. To build your own project on Hive's agent loop and models, depend onaden-hive: see Building on Hive.
Then type a task on the home screen and choose the queen to give it to, or open the Prompt Library and send a ready-made prompt straight to the queen it was written for.
How it works
flowchart LR
You(["You"]) -->|"describe the outcome"| Queen["Queen<br/>(persistent agent)"]
Queen -->|"proposes a colony,<br/>you approve"| Pilot["Pilot<br/>(one unit, done by the queen)"]
Pilot -->|"saves the method"| Skill["Skill + playbook"]
Skill -->|"run_worker / run_playbook"| W["Worker clones<br/>in parallel"]
W -->|"tracker_upsert"| T[("Tracker<br/>shared SQLite")]
T -->|"SQL: what's done,<br/>what's left"| Queen
Queen -->|"checked answer"| You
style Queen fill:#ffb100,stroke:#cc5d00,color:#333
style T fill:#fff3d6,stroke:#cc5d00,color:#333
style W fill:#ff9800,stroke:#cc5d00,color:#fffHive has one execution primitive: the agent loop. The queen is one, and every worker is a clone of it with its own task, a narrower set of tools and a strict budget. Orchestration is a tool call, not a compiled graph:
run_workerfans tasks out and returns at once, so the queen keeps talking to you while workers run. Four run at a time by default; the rest wait their turn. Each worker's report arrives in the queen's conversation as a new turn.The tracker is the colony's shared state. The queen defines the table and which columns workers may write, workers upsert one row per unit of work, and the queen checks progress with SQL. It's a file on disk:
~/.hive/colonies//tracker/tracker.db.run_playbookruns a proven method over every row, with retries and backoff, rate-limited lanes, and a dead-letter list for rows that keep failing. "What's left" is always a fresh query of the tracker, so re-running a playbook picks up where it stopped.
The architecture overview goes deeper: the loop, the tool surface, memory, human oversight, and how state survives a crash.
Home. Your queens and their colonies on one map. Describe a task and choose who takes it. Workers. One per unit of work, each with its own task and budget, all reporting to the queen. Tracker. Results land in a shared table as workers finish, ready to query, export or resume from. Result. A checked answer with sources and a chart, right in the chat.
What's inside
Queens with a job and a memory. Hive ships thirteen persona queens. Six are active out of the box (Growth, RevOps, Content, Lead Generation, Outbound, and Brand & Design); hire the rest from the Org Chart, or create your own. Each one keeps notes in markdown memory as she works, and relevant past conversations are recalled into her context automatically.
A memory you can measure. A queen remembers you across sessions. She writes notes about you as she works, related past conversations are placed in front of her before each message, and she checks her memory before asking you for something you've already told her. We benchmark it on LongMemEval, run end to end through Hive and graded by the official judge, with gpt-6.1-sol answering:
| LongMemEval-S | LongMemEval-M | |
|---|---|---|
| History per question | ~115k tokens | ~1.1M tokens |
| Hive | 88.4% | 82.0% |
| The same model reading the whole history | 96.8% | doesn't fit |
When a whole history fits in the model's context, reading all of it is still more accurate than Hive's memory. When it doesn't, as on LongMemEval-M, memory is the only option. The benchmark page has the methodology, per-type results, a comparison with other memory systems, and the limitations.
Built-in tools that run in-process. Shell commands and background jobs, file editing, fast code search, PDFs, attachments and images, web scraping, charts (ECharts and Mermaid), CSV files, and image generation through Hive LLM. They run inside Hive itself, with no tool servers to start.
Your browser, driven by your agents. The Hive Browser Bridge extension lets agents work in your own Chrome, where you're already logged in. Each worker gets its own tab group.
Lo que la gente pregunta sobre hive
¿Qué es aden-hive/hive?
aden-hive/hive es subagents para el ecosistema de Claude AI. Multi-Agent Harness for Production AI Tiene 11.1k estrellas en GitHub y su última actualización registrada es del 2026-10-19.
¿Cómo se instala hive?
Puedes instalar hive clonando el repositorio (https://github.com/aden-hive/hive) o siguiendo las instrucciones del README en GitHub. ClaudeWave también te ofrece bloques de instalación rápida en esta misma página.
¿Es seguro usar aden-hive/hive?
Nuestro agente de seguridad ha analizado aden-hive/hive y le ha asignado un Trust Score de 100/100 (tier: Verified). Revisa el desglose completo de comprobaciones superadas y flags en esta página.
¿Quién mantiene aden-hive/hive?
aden-hive/hive es mantenido por aden-hive. La última actividad registrada en GitHub es del 2026-10-19, con 1093 issues abiertos.
¿Hay alternativas a hive?
Sí. En ClaudeWave puedes explorar subagents similares en /categories/agents, ordenados por popularidad o actividad reciente.
¿Mantienes este repo? Muestra su Trust Score en tu README
La insignia muestra el Trust Score y el veredicto actuales de este repositorio y enlaza a esta página. Sigue los datos del catálogo: nadie puede comprar ni editar la puntuación que enseña.
[](https://claudewave.com/en/repo/aden-hive-hive)<a href="https://claudewave.com/en/repo/aden-hive-hive"><img src="https://claudewave.com/api/badge/aden-hive-hive" alt="ClaudeWave Trust Score for aden-hive/hive" width="320" height="64"></a>