↓ Skip to main content
tomrochette.com
  1. Agents/
  2. Orchestration/

Hive

Author
glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 5 minutes read (~812 words)
agent-curated 418 fully-ai-generated 520 llm=glm-5.3-flash 429 multi-agent 33 orchestration 68 python 28 research-note 388
Table of Contents

Hive is Aden’s Apache-2.0 Python runtime for “colonies” of agents: a persistent Queen lead that spawns worker clones of itself around a shared tracker ledger and a persistent plan, positioned for business-process automation rather than coding sessions.

Its bet is that coordination needs one primitive, not a graph: the Queen is an agent loop and every worker is a clone of it, so there is no orchestration DSL to learn and the colony grows at runtime.

What it is
#

A zero-setup, model-agnostic runtime (OpenAI, Anthropic, and Gemini are the documented providers) where you describe an outcome in natural language and the Queen grows a colony of specialized clones to run it, coordinating through a shared ledger and a persistent plan with crash-safe park/resume and cost enforcement. Human oversight is built in as Sentinel, an out-of-band human-in-the-loop channel, alongside three-level observability. Surfaces are a CLI (hive run, hive tui, hive deploy), Docker self-host, and Aden Cloud for managed deployment, with a documented Claude Code integration (/hive) for building the agents. It is made by Aden, a five-person San Francisco company from Y Combinator’s Winter 2020 batch that previously built the Acho data-worker app.

Status
#

Active but uneven: 11,086 stars as of 2026-10-08 on a repository created 2026-01-12, with branches pushed the day before this check (GitHub API). The default branch’s last commit is 2026-09-14 and the GitHub release line stopped at v0.11.0 on 2026-05-02, five months before this check, so the code ships without tagged releases. Three contributors dominate the history (about 1,179, 868, and 451 contributions), which matches a five-person company rather than a community.

Star History Chart

The independent footprint is near zero: no Hacker News thread about it cleared 3 points through 2026-10-08, and the sharpest record is a 1-point, zero-comment December 2025 thread accusing Hive of taking Beads’ approach to agent memory, unanswered and uncorroborated. The name also collides with Apache Hive, Hive OS, and a smart-home product line, so every search for it is polluted.

Strengths
#

  • One coordination primitive: workers are clones of the Queen, so new colony members inherit the same tools and model without wiring a graph.
  • The production plumbing is real in the docs: crash-safe park/resume, per-colony cost enforcement, three-level observability, and Sentinel human approval.
  • Apache-2.0 with Docker self-host or Aden Cloud deploy, and a Claude Code path for building the agents with the harness you already run.

Cautions
#

  • The release train stopped at v0.11.0 in May 2026 while trunk commits run monthly, the same ship-without-releasing pattern this category flagged on AgentsMesh.
  • Three contributors dominate the history, so this is a single company’s product with no community governance.
  • Naming costs: the repo is hive, the product is OpenHive or Aden Hive, the company is Aden, and three unrelated Hives own the search results.
  • The adoption record is the star count alone: no independent field reports, and an unresolved December 2025 provenance complaint from the Beads camp sits in the record.

Pricing
#

No public prices: the runtime is free and Apache-2.0, and Aden Cloud is the hosted deploy without a published price table as of 2026-10-08. No price history table until a price is published.

Compared to
#

  • CAMEL: the research framework for studying agent societies; choose CAMEL to study colonies, Hive to operate one.
  • AG2: the AutoGen community fork’s conversational framework; choose AG2 for community-governed Python conversations, Hive for a ledger-and-clones runtime with crash recovery.
  • Raven: the DAG-planning host over coding agents; Raven plans engineering tasks, Hive’s colonies run business processes.

Bottom line
#

Recommended for teams automating long-running business processes who accept a young, company-driven runtime with a stalled release train. Not for coding-session orchestration, where the parallel-agent columns are stronger, or for anyone who needs an active release line, community governance, or independent validation. My disagreeable claim: the 11.1k stars measure the colony pitch, not the artifact, because five months without a release, three dominant contributors, and near-zero independent discussion mean the count is riding the multi-agent wave rather than adoption, and I expect it to decouple from usage unless the release train restarts.

Changes
#

  • 2026-10-08 - Created from the kaushikb11/awesome-llm-agents entrant scan.

See also
#

  • CAMEL - the research wing of the framework family this note joins
  • AG2 - the community-fork framework alternative in the same wing
  • Raven - the DAG-planning host agent with its own launch controversy
  • PraisonAI - the low-code multi-agent framework closest in packaging
  • Orchestration Feature Matrix - the category comparison this note joins

References
#