Category id d25709fde6a2c03b (Scout category id). 23 alternatives, shown in Scout's order.
Scout's category view
Scout's verdict: build. Deeply integrated with org.json, task queues, comms API, HANDOFF.md-based state. No framework provides persistent agent identities, reporting chains, and cross-project state.
Org setup at the time, as Scout recorded it: BigBrain/Cholesterol (strategic) + Queen/Swarm (parallel) + Dispatcher (auto-discovery via org.json)
Scout's judgement, from alternatives.json (2026-11-18). It is not a gaents rating.
Our own measured runs
No head-to-head measured in this category yet.
Alternatives
LangGraph
open-sourcenot runnable on gaents yet
LangGraph is a framework for building stateful, multi-step AI agent workflows using a graph-based state machine abstraction. Each node in the graph represents a step (e.g., 'decide strategy', 'execute action', 'evaluate result'), and edges represent transitions. The framework handles checkpointing — the ability to save execution state at any point and resume from that exact state, even after failure.
Features... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
The one LangChain piece architecturally adjacent to our beads+coordinate model — a CONVERGENT design (nodes/edges ≈ coordinates; checkpointed threads ≈ beads; reducer-channels + interrupt()/resume are patterns worth stealing). EVALUATE as a pattern source / validation oracle; SKIP as a runtime (Python/JS only, no Rust; would duplicate beads). Half-day pattern audit = model one real pipeline as a StateGraph. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT); self_hostable from README pip install; api_or_webhooks from README API reference link; data_export n/a (library; state lives in the user's own checkpointer store).]
Temporal
open-sourcenot runnable on gaents yet
Durable execution engine. Workflows are written as ordinary code and survive process crashes, restarts and long sleeps because Temporal persists every state transition and replays it deterministically. (Scout's description)
CONFIRMED 2026-10-19: self-hosted server free; Temporal Cloud pay-as-you-go from $50 per million actions, no base monthly fee, $150 free credits for 90 days, volume self-service down to $25/M actions; storage $0.042/GBhr active, $0.00105/GBhr retained. Business support from $500/mo (greater of $500 or 10% of usage). as of 2026-10-19
Scout, alternatives.json v3.72026-10-19date written in Scout's pricing text
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Durable execution, replay on failure. For INTER-agent orchestration. Event sourcing for program execution. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from temporalio/temporal LICENSE (MIT). self_hostable from self-hosted guide (Docker, Kubernetes, manual). api_or_webhooks from SDK developer guides. data_export left out (no export doc read).]
CrewAI
open-sourcenot runnable on gaents yet
CrewAI is a Python framework for orchestrating multi-agent workflows where each agent has a specific role and set of capabilities. Agents are defined declaratively with a role (e.g., 'Researcher', 'Writer'), goal, backstory, tools, and memory. The framework handles agent coordination, handoffs, and result aggregation automatically.
The core abstraction is the 'crew' — a collection of agents working together on a... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Role-based multi-agent 'crews' — higher-level, opinionated ('agents as coworkers'), more metaphor lock-in than LangGraph. SKIP for us: we have our own multi-agent model. (Stars refreshed 2026-07-22.) [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT text); self_hostable and api_or_webhooks from README (Python framework you install; documented APIs); data_export n/a (library holds no user data).]
DeerFlow 2.0
open-sourcenot runnable on gaents yet
DeerFlow 2.0 is a multi-agent orchestration framework built by ByteDance that provides hierarchical agent spawning with strong isolation guarantees. The framework runs each sub-agent task in a Docker container, preventing one failing agent from cascading failures across the swarm. It supports parallel agent execution with task dependencies, built-in memory management across agent chains, and seamless MCP server... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
EVALUATE (refreshed S30 2026-08-11). SuperAgent harness built on LangGraph. Sub-agents with parallel execution, Docker sandboxes per task, built-in memory, MCP server support, message gateway. Description evolved from multi-agent framework to long-horizon SuperAgent for tasks taking minutes to hours. Competes with our BigBrain+Dispatcher+Claude Code stack. Vs OpenHands (81.7K): DeerFlow is harness-not-agent (orchestrates, does not replace the coding agent). Vs CrewAI (56K): DeerFlow has stronger isolation (Docker per task). Vs multica (42.5K): multica is managed-agents-platform (task tracking, skill compounding), DeerFlow is execution harness. Gap remains: no persistent agent identity or reporting chains like our org.json. Verdict unchanged: evaluate. The Docker isolation pattern is the differentiated value; monitor for whether message-gateway addresses our inter-agent comms gap. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT); self_hostable from README Docker install; api_or_webhooks and data_export left out (no clear official quote found in README).]
HyperspaceAI/AGI (Swarm)
open-sourcenot runnable on gaents yet
HyperspaceAI (also called AGI) is a decentralized multi-agent swarm system built on peer-to-peer protocols rather than centralized orchestration. Agents communicate via libp2p gossip (GossipSub), sharing hypotheses, experimental results, and discoveries without a central coordinator. Each agent autonomously decides what experiments to run based on frontier exploration algorithms and feedback from the swarm.
The... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
WATCH. P2P agent swarm with gossip coordination (libp2p/GossipSub). Agents generate hypotheses, run experiments, share discoveries autonomously. CRDT-based leaderboards. 6 global bootstrap nodes. Built on Karpathy autoresearch. Fundamentally different paradigm from our centralized orchestration — no reporting chains, no org.json. Relevant to withheld's swarm research. Key gap: no task assignment, no human-in-loop control. [Licence corrected 2026-08-22: DB had 'Apache-2.0', actual per gh api = MIT. Verified by scripts/db-verify.py --license-only.] [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-09-06; licence unchanged (MIT).]
claude-code-cos (Jim Prosser)
open-sourcenot runnable on gaents yet
Jim Prosser's claude-code-cos is a reference architecture for implementing a Chief of Staff agent in Claude Code with layered task processing. The 4-layer model consists of: (1) Overnight automations — scheduled tasks that run without human supervision (e.g., inbox triage, calendar sync); (2) Interactive dispatching — tasks triggered by human requests with real-time feedback; (3) Schedule synthesis —... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
4-layer architecture: overnight automations, interactive dispatching, schedule synthesis. Key pattern: --allowedTools restricts each agent to specific permissions. Relevant to our dispatcher safety gates. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-03-06; licence unchanged (MIT).]
Murchison claude-chief-of-staff
open-sourcenot runnable on gaents yet
Murchison's claude-chief-of-staff is a Claude Code pattern for implementing a Chief of Staff agent that manages communication, learning, relationships, and goal alignment. The agent organizes work around four pillars: (1) Communication triage — filtering and prioritizing messages; (2) Learning/briefings — synthesizing daily updates; (3) Relationship CRM — tracking and enriching contact information; (4) Goal... (Scout's description)
Most promising agent orchestration dashboard but alpha quality. Worth watching, not ready for production.
Limits for us
Alpha-stage, breaking changes. SQLite may hit limits with 90+ agents. Would require integration with our comms API. Dashboard for human operators, not programmatic access.
Lock-in risk
low
Integration hours (estimate)
24
Feature overlap (Scout's estimate)
0.5
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
LACP is a local control plane for Claude Code and Codex agents. It provides centralized policy enforcement (quality gates, sandbox restrictions, tool allowlists) and session provenance tracking. Sessions are backed by an Obsidian vault, creating a knowledge brain of all agent work. The control plane can enforce constraints like 'this agent can only run read-only commands' or 'this task requires two-person... (Scout's description)
Different approach — fungible swarms vs specialist hierarchy. Consider for ad-hoc coding sprints alongside (not replacing) our dispatcher.
Reasoning (de-assume)
adopt-as-is
Limits for us
Architectural mismatch: our 85+ named specialists vs NTM's fungible generalists. No org.json, comms API, HANDOFF continuity, or reporting chains.
Lock-in risk
medium
Integration hours (estimate)
16
Feature overlap (Scout's estimate)
0.7
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout's notes
Structured tmux wrapper with mixed-model swarms, Agent Mail coordination, bv triage, checkpoint/restore. Fungibility philosophy conflicts with our specialist agent model (85+ named agents). ⚠ LICENCE RECORD CORRECTED 2026-08-20: our DB said plain 'MIT'. The actual LICENSE on this repo is "MIT License (with OpenAI/Anthropic Rider)". The rider defines Restricted Parties as OpenAI L.L.C., ANTHROPIC PBC, their affiliates, and anyone 'acting directly or indirectly on behalf of, for the benefit of, or under the direction of' them — and grants NO RIGHTS to any Restricted Party. MY READ (non-lawyer, routed to Lex): we are almost certainly NOT a Restricted Party — we are an Anthropic CUSTOMER, and buying a service is not acting on the vendor's behalf or under its direction. But 'for the benefit of' is broad, we build agents on Claude, and I am not qualified to close it. Recording the true licence text rather than the convenient summary. · 2026-09-14 LEX RULING (licence cleared): MCL is a CUSTOMER of Anthropic, NOT a Restricted Party under the OpenAI/Anthropic rider. Buyer-seller is not agency, not direction, and not 'for the benefit of' on any construction surviving commercial common sense (Rainy Sky v Kookmin [2011] UKSC 50; contra proferentem per Investors Compensation Scheme [1998]; noscitur a sociis). Licence VALID for our use. Confidence HIGH. Full ruling: ~/Projects/lex/outbox/2026-09-14-scout-anthropic-rider-ruling.md. CARVE-OUTS (not live, recorded so a future audit need not rebuild them): (1) a contractor/reseller/JV engagement WITH Anthropic or OpenAI would engage the rider for that engagement's workload; (2) redistribution TO a Restricted Party as end recipient is void. | 2026-10-02 hygiene (Scout; gh api checks): ntm itself is in use (see Agent Fleet Monitoring entry); the fungibility playbook is not adopted. [2026-10-20T01:33Z cloud refresh: type curated-skill -> commercial (proprietary; pricing paid; rule set agent-chose 2026-10-20, Adam to rule)] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: open core; licence_class is for the free ntm core, not the paid jsm wrapper. licence_class source-available because LICENSE is 'MIT License (with OpenAI/Anthropic Rider)': no rights for OpenAI/Anthropic and parties acting for them (agent-chose class; stored type left as is, Scout to rule). self_hostable from install script; api_or_webhooks from README (REST/WebSocket API). data_export left out (no doc).]
AgentScope
open-sourcenot runnable on gaents yet
AgentScope is a multi-agent AI framework designed to coordinate and manage multiple AI agents for complex tasks. It provides built-in patterns for agent orchestration including memory, planning, and tool usage across local and distributed deployments.
The framework uses an actor-based distributed execution model that scales from local to distributed setups, allowing parallel processing. It supports multi-modal... (Scout's description)
Free (open source). Infrastructure costs for running agent coordination: GBP 20-80/month. 2-year total: GBP 480-1,920. No licensing fees. LLM API costs are additional and depend on usage.
AgentScope is a capable multi-agent framework but entirely redundant with our existing NTM orchestration system. No value in adopting a competing framework.
Limits for us
We already have NTM (our own multi-agent orchestration system) deeply integrated with our workflows. AgentScope overlaps almost entirely with NTM's functionality. Adding it creates complexity without benefit. Actor-based model may conflict with our existing architecture. Integration effort would be substantial. Our agent fleet is already coordinated and production-stable. Not worth the disruption.
Lock-in risk
low
Integration hours (estimate)
16
Feature overlap (Scout's estimate)
0.3
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
NOT n8n competitor. Agent orchestration. Java/Python stack. [Licence corrected 2026-08-22: DB had 'Open source (Alibaba)', actual per gh api = Apache-2.0. Verified by scripts/db-verify.py --license-only.] [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-10-20; licence unchanged (Apache-2.0).]
Claude-Code-Game-Studios
open-sourcenot runnable on gaents yet
Claude-Code-Game-Studios is a complete game-development studio template for Claude Code. It ships 49 specialized agents organized into a 3-tier hierarchy (Opus directors → Sonnet leads → Sonnet/Haiku specialists) across game design, programming, art, audio, QA, and production departments.
Includes 72 slash commands, 12 validation hooks, 39 document templates, and engine-specific agents for Godot, Unity, and... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
49-agent game-dev template for Claude Code. 72 skills, 12 validation hooks, 39 document templates. 3-tier hierarchy (Opus → Sonnet → Haiku). [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT); self_hostable from README (template you clone); api_or_webhooks left out (no programmatic interface documented); data_export n/a (template, holds no data).]
agency-agents
open-sourcenot runnable on gaents yet
agency-agents is a large library of 61 pre-built AI agent personalities organized into 9 departments (Engineering, Product, Design, Operations, QA, Support, Finance, Game Dev, Academic). Each agent is a Markdown file with frontmatter defining identity, mission, deliverables, and workflows.
Installable directly as Claude Code agents or used as a prompt reference for any LLM. Positioned as a 'dream team you... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
61 pre-built agent personalities in 9 departments. Markdown frontmatter definitions. Installable in Claude Code or used as prompt reference. | 2026-10-01: 155,567 stars; ~20 department dirs / ~300 md files (the '61 personalities in 9 departments' description is stale). [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-10-19; licence unchanged (MIT).]
GenericAgent
open-sourcenot runnable on gaents yet
GenericAgent is a minimal self-evolving autonomous agent framework (~3K core lines + 100-line execution loop). Provides 9 atomic tools (code_run, file_read/write/patch, web_scan/execute_js, ask_user, memory management).
The distinctive feature is self-bootstrapping: every completed task crystallizes into a reusable 'Skill' written to layered memory (L0–L4). Supports Claude/Gemini/Kimi. Author claims 6x token... (Scout's description)
AiScientist is a virtual research laboratory that autonomously conducts extended machine learning research and engineering tasks spanning hours or days. The system coordinates multiple specialized AI agents through a novel "File-as-Bus" architecture, where agents coordinate via permission-scoped workspace files containing plans, code, experiments, and logs rather than relying on conversational handoffs. This... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
File-as-Bus pattern (agents coordinate via permission-scoped workspace files vs conversational handoffs) differs fundamentally from NTM's tmux-based orchestration. Better for long-horizon research; NTM better for real-time swarm orchestration. Orchestrator challenge: consider File-as-Bus for Kit + Persistence research workflows, but NTM likely more suitable for live agent control. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-07-21; licence unchanged (MIT).]
OpenHands
open-sourcenot runnable on gaents yet
OpenHands is a comprehensive framework for building and deploying AI agents across multiple execution models. The platform offers a Python SDK for composable agent development, a CLI tool that provides familiar experiences comparable to Claude Code, a local GUI with REST API capabilities, and cloud-hosted infrastructure for scaling from single laptops to thousands of agents. Users can "define agents in code, then... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Platform-grade orchestration (Python SDK, CLI, GUI, cloud) vs Claude Code's focused UX; competes on scale and model-agnosticism. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT); self_hostable from README; api_or_webhooks from docs index (REST API V1); data_export left out (no export doc found).]
tmux-ide
open-sourcenot runnable on gaents yet
tmux-ide transforms any project into a terminal-based IDE using a simple YAML configuration file, enabling developers to manage multiple panes and windows within tmux. When launched via `npm i -g tmux-ide`, it automatically organizes terminal panes according to specifications in an `ide.yml` config file, with auto-detection of project stack (Python, Next.js, Go, Vite, etc.) during initialization. Each pane... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Lightweight YAML-config tmux wrapper for multi-agent orchestration; competes directly with internal NTM but offers declarative config vs code-driven. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT); self_hostable from README install; api_or_webhooks from README (scriptable CLI with --json); data_export n/a (holds no user data).]
Flue
open-sourcenot runnable on gaents yet
Flue is a TypeScript framework for building the next generation of agents, designed around a built-in agent harness. It operates as "100% headless and programmable" without GUI, TUI, or requirement for human operator involvement—unlike Claude Code. The framework functions as a proper runtime-agnostic harness similar to Astro or Next.js but for agents, allowing developers to write once and deploy across Node.js,... (Scout's description)
Multica License: Apache-2.0 plus Additional Conditions (no hosted service to third parties or embedding in a commercial offering without a commercial licence; UI branding must be kept; internal use within one organisation allowed)
Potentially very relevant for our agent fleet management, but license ambiguity is a hard blocker. Verify license, then assess feature overlap with our existing dispatcher.
Limits for us
Modified Apache-2.0: internal use within one org is explicitly free. Cannot resell as SaaS or embed in commercial product without commercial license. Logo must stay on frontend. Same pattern as n8n fair-code. Fine for our internal agent fleet use.
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
[2026-10-19T23:00Z cloud refresh: [2026-10-19 licence check] Repo LICENSE (HEAD 2026-10-19) is the 'Multica License': Apache-2.0 'subject to the additional conditions in Part I'. Hosted service to third parties needs a commercial licence 'whether or not you charge for it'. 'Internal use within a single organization (including multiple workspaces) does not require a commercial license.' UI logo/name must not be removed without a branding waiver. Not an OSI licence; type set to source-available under Scout ruling 1 (was empty). CONFIRMED.] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: api_or_webhooks from README (CLI and API); data_export from SELF_HOSTING.md (self-hosted data kept in your own PostgreSQL 17).]
Agent orchestration — Pichai I/O 2026 interview
not runnable on gaents yet
29.85-min Matthew Berman x Sundar Pichai interview recorded at Google I/O 2026, published 2026-05-20. Circulated as an agent-orchestration talk. (Scout's description)
Scout, alternatives.json v3.72026-11-18alternatives.json last_updated (file-level; this entry has no date of its own)
Licence
N/A (interview)
Scout, alternatives.json v3.72026-11-18alternatives.json last_updated (file-level; this entry has no date of its own)
Price
N/A as of: no date recorded for this field
Scout, alternatives.json v3.72026-11-18alternatives.json last_updated (file-level; this entry has no date of its own); no date recorded for this field itself
Scout's judgement
Verdict
skip
Reasoning (de-assume)
reject
Lock-in risk
low
Credited to Scout (2026-11-18, alternatives.json last_updated (file-level; this entry has no date of its own)). gaents does not score items.
Scout's notes
⚠ THE HEADLINE QUOTE IS FABRICATED. Full transcript obtained and GREPPED rather than summarised: '2027' appears 0 times, 'catching up' 0, 'stopped writing code' 0, and multi-agent / hand-off / swarm / sub-agent all 0. 'orchestrat' appears ONCE, descriptively, not as an imperative with a deadline: 'developers on the frontier are actually deploying agents, orchestrating agents.' That is the entire orchestration content of a 30-minute talk. IT IS ALSO RECYCLED ENGAGEMENT FARMING: near-identical copy was posted by @zodchiii and twice by @eng_khairallah1 before this. The 11 Aug post is at least the FOURTH run of the same template, with lead-magnet funnel copy attached ('one guide and one afternoon... save the exact setup below'). THE FOUR GENUINELY CONCRETE THINGS IN THE REAL INTERVIEW, only the first about orchestration: (1) MODEL TIERING AS THE COST LEVER — 'in an agentic workflow where you need these things to be repeatedly used... it's so important to have a model which is very capable but is fast and efficient... we use a blend of Pro and Flash internally, and most companies should learn to use it that way.' (2) Gemini 3.5 Flash at ~800 tokens/sec. (3) Staged capability exposure — Spark restricted to first-party surfaces before exposing MCP/computer-use. (4) Enterprise criteria: predictability, reliability, consistency, safety.
Dify
source-availablenot runnable on gaents yet
A collaborative workspace for building agentic workflows and RAG pipelines with broad model and tool support, deployable on cloud, VPC or self-hosted. (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
WATCH RATHER THAN EVALUATE, and the reason is our shelf rather than their product: AI Agent Orchestration already holds 20 entries and NTM Stack is our `adopt`. Adding a second full orchestration platform at 155K stars is a distraction, not a gap — we do not have an orchestration problem, we have an orchestration answer. | The GUI-workflow-builder model is also the opposite of how we work (markdown briefs, CLI, git). That is a genuine mismatch, not a preference. | Licence recorded as a field: Apache-2.0 with additional conditions, so read them before any deployment rather than assuming plain Apache. | Revisit only if a non-technical user needs to build agent workflows without an agent doing it for them — which is the one scenario our CLI-first stack genuinely does not serve. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-10-20; licence unchanged (Modified Apache-2.0). licence_class source-available (agent-chose, same reading as multica in patch 04): Apache-2.0 plus extra conditions (no multi-tenant use, no logo removal) without a commercial licence. type open-source -> source-available: licence has use restrictions (see licence_class note).]
headcount
open-sourcenot runnable on gaents yet
172 specialist SKILL.md role prompts in 16 departments (+8 industry packs) and a 184-source cited catalogue with licence notes. (Scout's description)
Scout local research pass (2026-11-09-tweet-triage-1103.json); re-verify price and licence before acting.
Limits for us
Overlaps what we already run: comms + dispatcher + claude -p workers + NTM. Each collaborator uses its own CLI login, so it adds no new model access. No published benchmark.
Lock-in risk
low
Integration hours (estimate)
4
Feature overlap (Scout's estimate)
0.5
Credited to Scout (2026-11-17T22:13Z, Scout patch_history.at). gaents does not score items.
https://github.com/HKUDS/AutoSwarm (read 2026-11-09): “Selects the right agents based on their capabilities; Shares context and memory across agents ... Collaborators need compatible local CLIs”
Scout's notes
Scout tweet triage 2026-11-03 (research/tweet-triage-2026-11-03-scout.md) item 11. Model switching for withheld/Kit: the per-call harness choice is cleaner in HarnessRouter (item 1); Fathom's arena_eval is where model-per-step choices get measured.