Agents

• LangGraphStateful multi-actor agent workflows

  Free / OSS · OSS/MIT · 43.6k GitHub stars

  VerdictThe right tool when agents need memory and branching logic.

  CatchThe abstraction overhead is real. Debugging graph execution requires understanding the underlying state management, and simple agents become verbose.

• CrewAIRole-playing agents that actually ship

  Free / OSS · OSS/MIT · 59.9k GitHub stars

  VerdictBest DX for multi-agent systems if you embrace the paradigm.

  CatchThe simplicity cuts both ways. Complex inter-agent communication patterns require workarounds, and you are locked into CrewAI execution model.

• HatchetMIT durable tasks/agents on Postgres; Cloud free tier real

  Free / OSS · Freemium · 8.2k GitHub stars

  VerdictMIT durable tasks on Postgres. Cloud free tier real; self-host forever.

  CatchVs Temporal/Inngest/Trigger.dev: Postgres-backed durable tasks + agent runs — not Temporal’s own cluster story, not Inngest/Trigger event-function DX alone. Cloud paid past free run allotment. Self-host is $0 software.

• DeepEvalApache-2.0 pytest-style LLM/agent evals; Confident AI is the paid cloud

  Free / OSS · Apache-2.0 · 19k GitHub stars

  VerdictApache-2.0 LLM unit tests. Free locally; Confident AI starts free then $200.

  CatchOSS is the runner — collaboration/dashboards/monitoring are Confident AI (Free: 2 seats/1 project/5 runs/wk; Starter $200/mo; Team $2k/mo). LLM-as-judge burns provider tokens. Vs Ragas: broader agent/chat coverage, not RAG-only.

• smolagentsApache-2.0 minimal code-agent framework from Hugging Face

  Free / OSS · Apache-2.0 · 30k GitHub stars

  VerdictApache-2.0 code-first agents. Free library; sandbox/ops are yours.

  CatchMinimalist — not Temporal durable workflows or production multi-tenant agent platform. Code execution needs sandbox discipline. Vs CrewAI/PydanticAI: HF Hub gravity, less enterprise packaging. You own evals, memory, and deploy.

• AgnoApache-2.0 Python agent framework + AgentOS runtime (ex-Phidata)

  Free / OSS · Apache-2.0 · 42.8k GitHub stars

  VerdictApache-2.0 Python agents + AgentOS. Ship on Agno, not Phidata nostalgia.

  CatchVs LangGraph (graphs), CrewAI (roles), Mastra (TypeScript): Agno is Python performance + AgentOS productization, not a coding CLI like Claude Code/OpenCode. Phidata docs/imports are dead—use Agno. Control Plane and hosted conveniences are optional; day-one is OSS SDK+runtime you operate.

• SteelApache-2.0 browser API for AI agents — self-host or cloud

  Free / OSS · Apache-2.0 · 7.8k GitHub stars

  VerdictApache-2.0 agent browser API. Self-host Steel; pay Browserbase-shaped cloud only if you must.

  CatchVs Browserbase (managed fleet), Stagehand (AI Playwright layer), Browser Use (LLM browser framework): Steel is the browser sandbox/API — bring your own agent logic. Self-host means Chrome resource cost; cloud is freemium/paid convenience. Anti-bot arms race still applies.

• Google ADKApache-2.0 code-first agent build/eval/deploy kit (Google)

  Free / OSS · Apache-2.0 · 21.9k GitHub stars

  VerdictApache-2.0 Google agent kit. Code-first build/eval/deploy; Vertex optional, not mandatory.

  CatchVs Gemini CLI / Antigravity CLI (terminal coding agents), CrewAI/LangGraph/Agno (orchestration frameworks), and Daytona/E2B (sandboxes): ADK is the Google build/eval/deploy kit, not a coding CLI and not a sandbox fleet. Vertex/GCP gravity is real for enterprise deploy; OSS still runs without a Google Cloud bill if you stay local/BYO.

• Spec KitMIT Spec-Driven Development toolkit — any AI coding agent (GitHub)

  Free / OSS · OSS/MIT · 142k GitHub stars

  VerdictMIT SDD toolkit. Specs before code; not a coding agent itself.

  CatchVs Claude Code / Codex / OpenHands (coding agents): Spec Kit is the SDD workflow primitive — process and toolkit, not a coding agent seat. You still bring the agent, models, and repo discipline. GitHub org project, not a hosted product SLA.

• Google Workspace CLIApache-2.0 first-party gws CLI — Drive, Gmail, Calendar, agent skills

  Free / OSS · Apache-2.0 · 31.7k GitHub stars

  VerdictApache-2.0 Workspace CLI. First-party org tooling for agents; not a Google support SLA.

  CatchFirst-party googleworkspace org surface — but the README states it is not an officially supported Google product. Vs raw Workspace REST / Apps Script: this is the CLI/agent skill layer, not a new API. Needs a GCP project + OAuth; unverified apps hit scope limits; Workspace/Admin API quotas still apply.

• OpenResearchMIT coding agents → research agents (Rust)

  Free / OSS · OSS/MIT · 7.4k GitHub stars

  VerdictMIT research-agent workspace. Coding agents → autoresearch; not Spec Kit/DSPy/Promptfoo.

  CatchVs Spec Kit (spec-driven coding workflow), DSPy (prompt/program optimizers), Promptfoo (eval/red-team): OpenResearch is a research-agent harness and experiment ledger around existing coding agents — not a spec kit, not a prompt compiler, not an eval suite. You still pay (or self-host) the coding agent and any remote compute you dispatch.

• Atlas (pacifio)MIT multi-agent change tracking — source control for agents

  Free / OSS · Apache-2.0 · 9.6k GitHub stars

  VerdictMIT agent source-control. pacifio Atlas ≠ Ariga Atlas; track agents, not schemas.

  CatchVs Worktrunk/jj (git worktree / change-oriented VCS plumbing): Atlas is agent session+change tracking with shared memory, not a general VCS replacement. name collision with Ariga Atlas (slug `atlas`, schema-as-code) — this card is `atlas-agents` / pacifio only; never confuse with atlasgo.io.

• DeskcommCRMMIT self-hosted AI sales OS/CRM — MCP-ready

  Free / OSS · OSS/MIT · 4.7k GitHub stars

  VerdictMIT self-hosted AI sales CRM. Agents+WhatsApp vs Chatwoot/Composio; HostGator kit ≠ lock.

  CatchVs Chatwoot (open inbox/helpdesk) and Composio (agent tool auth/connectors): DeskcommCRM is an AI sales OS/CRM with WhatsApp agents, not a general shared inbox alone and not a universal tool marketplace. HostGator affiliate/setup kit is infra convenience, not product lock — you can host elsewhere.

• OpenSpecMIT spec-driven change tooling for coding agents

  Free / OSS · OSS/MIT · 72.6k GitHub stars

  VerdictMIT spec-driven agent tooling. Align before code; host agent required — not a product alone.

  CatchNeeds an AI coding-agent host — OpenSpec is tooling around the agent, not a product that ships code by itself. Vs Spec Kit (heavier phase-gated SDD): OpenSpec is lighter and more iterate-friendly, still not a standalone SaaS. You still pay (or self-host) the coding agent that runs the slash commands.

• ECCMIT agent ecosystem / coordination surface

  Free / OSS · OSS/MIT · 285k GitHub stars

  VerdictMIT agent coordination ecosystem. Skills+memory+security on top of host agents — not a host itself.

  CatchStill depends on host agents (Claude Code / Codex / Cursor / etc.) — ECC coordinates and tunes the harness; it is not a replacement coding agent. Vs OpenSpec (spec-driven change tooling) and Thurbox (tmux multi-agent TUI): ECC is the skills/memory/security ecosystem pack, not a spec framework and not a pane orchestrator.

• BrowserSkillAttach coding agents to a real logged-in browser without interrupting the session

  Free / OSS · OSS/MIT · 9.3k GitHub stars

  VerdictMIT live-session browser bridge. The no-second-login wedge is real; inherited cookies and permissions are the Catch.

  CatchA live logged-in browser is also a live blast radius: the agent inherits cookies, tabs, permissions, and whatever account is open. Vs Stagehand, Browser Use, and Playwright MCP (separate browser instances or contexts): BrowserSkill trades isolation for continuity. Treat the attached session as the trust boundary, not as a sandbox.

• CoderDurable remote workspaces for humans and coding agents

  Free / OSS · AGPL-3.0 · 17.1k GitHub stars

  VerdictAGPL-3.0 durable remote workspaces for people and agents. Self-host is the free path; Premium/Enterprise are paid.

  CatchVs E2B and Daytona (ephemeral agent sandboxes): Coder is a durable workspace/control plane, not a disposable per-task sandbox. The free path is OSS self-host; Premium and Enterprise are paid product tiers. You still own cluster, identity, persistence, and upgrade operations on the OSS path.

• AX CheckHosted free agent-experience checklist + recorded coding-agent sessions

  Free / OSS · OSS/MIT

  VerdictFree hosted AX smoke test. Grades how agents use your product, not how good your model is.

  CatchGrades are provisional and technical-only — treat the report as a useful first pass, not a product-wide quality certificate. Vs Promptfoo / DeepEval: AX Check grades product AX (docs, CLI, MCP, SDK), not model quality, eval scores, or safety behavior. Hosted free is convenient; verify the session evidence and rerun after interface changes.

• OrcaParallel coding-agent ADE with worktrees, mobile control, and design mode

  Free / OSS · OSS/MIT · 100k GitHub stars

  VerdictMIT parallel coding-agent ADE. Use for worktree-centered agent operation; bring the CLIs, keys, and merge discipline.

  CatchVs AgentBox and Octop: Orca is an operator-facing coding-agent ADE with worktrees and design/mobile surfaces, not a generic sandbox/VM launcher or a self-hosted multi-user agent platform. BYO agent CLIs, credentials, model spend, and the operational discipline to reconcile parallel worktrees.

• T3 CodeRemote control surface for already-authenticated agent CLIs

  Free / OSS · OSS/MIT · 28.9k GitHub stars

  VerdictMIT remote control for authenticated agent CLIs. Convenient across web/desktop/mobile; the signed-in local process remains the trust boundary.

  CatchVs Orca and AgentBox: T3 Code is the remote-control surface for existing agent CLIs, not Orca's worktree/ADE workflow or AgentBox's sandboxed VM launcher. Authentication, local process lifetime, network exposure, and the permissions of the already-signed-in CLI remain your responsibility; a remote UI does not reduce that blast radius.

• FlueAstro TypeScript harness for autonomous sandbox agents

  Free / OSS · Apache-2.0 · 8.5k GitHub stars

  VerdictApache-2.0 Astro agent harness. Use when you want programmable TypeScript agents with real sandboxes; bring your own models and host.

  CatchVs AgentBox and Octop: Flue is a TypeScript framework for authoring agent harnesses, not a multi-user agent platform or a generic sandbox/VM launcher. You still bring models, credentials, and operational judgment; a dead flue.dev marketing domain is not required — the GitHub repo is the source of truth.

• CLI-AnythingApache hub that makes arbitrary software agent-native via CLI surfaces

  Free / OSS · Apache-2.0 · 55.7k GitHub stars

  VerdictApache agent-native CLI hub when the bottleneck is “the software has no agent API”; skip if you only needed browser control or a thin argv parser.

  CatchVs agent-browser and aclif: this is a CLI-surface generator/hub for arbitrary apps, not a browser automation agent and not a generic CLI framework alone. Many hub CLIs assume heavy upstream installs (GIMP, Blender, GIS, DAWs); star velocity is extreme (~50k) — budget the host apps, not only the thin CLI wrapper.

• DeerFlowMIT long-horizon SuperAgent harness — sandboxes, memory, skills, gateway

  Free / OSS · OSS/MIT · 84.2k GitHub stars

  VerdictMIT SuperAgent harness when you want sandboxes + skills + gateway in one OSS tree and will bring models plus Docker headroom.

  CatchVs OpenHands, LangGraph, and Mastra: DeerFlow is a full SuperAgent product harness with sandboxes/memory/gateway, not an open coding agent alone, a graph orchestration library, or a TypeScript workflow framework. BYO model API keys; Docker sizing is resource-heavy (multi-vCPU/GB guidance); star velocity is extreme (~83k).

• CubeSandboxApache concurrent lightweight agent sandbox — RustVMM/KVM, E2B-compatible

  Free / OSS · Apache-2.0 · 13k GitHub stars

  VerdictApache KVM sandbox when you need dense, E2B-shaped isolation on your own Linux hosts; confirm license text and hardware before betting the fleet.

  CatchVs E2B, Daytona, and Cloudflare Sandbox: this is a self-hosted RustVMM/KVM sandbox control plane, not a hosted sandbox SaaS, a workspace platform, or an edge isolate runtime. GitHub license detect shows NOASSERTION even though LICENSE is Apache-2.0 — verify the tree before compliance sign-off (a detection caveat). Linux + KVM on x86_64 (ARM64 maturing); ops weight is real.

• WhiteboardMIT desktop canvas where humans + coding agents share specs

  Free / OSS · OSS/MIT · 3.3k GitHub stars

  VerdictMIT human+agent canvas when specs and decisions need to live beside the coding agent — not another Excalidraw clone.

  CatchVs Excalidraw: this is a human+agent design/decision workspace wired to coding agents, not a general hand-drawn whiteboard. Extreme overnight star fill (~589) is discovery noise; a hosted team product is planned — keep the self-hostable OSS desktop as the free bar until that SKU is real.

• PaperclipMIT org control-plane for agent budgets, Board governance, heartbeats

  Free / OSS · OSS/MIT · 105k GitHub stars

  VerdictMIT org control-plane when agent budgets and Board governance matter more than another task board — verify install path and star-vs-age risk.

  CatchVs Multica: Multica is a coding-agent issue board; Paperclip is company OS / org control. Vs OpenClaw: messaging employee, not budgeted Board governance. ★85k, but repo created 2026-03 — maturity may not match stars; piped install.sh is a supply-chain note, not a free pass.

• RufloMIT meta-harness / execution swarm on coding CLIs (ex Claude Flow)

  Free / OSS · OSS/MIT · 78k GitHub stars

  VerdictMIT meta-harness when you need an execution swarm on existing coding CLIs — not a substitute for org governance or a simpler single-agent CLI.

  CatchVs Pi / OpenCode / DeepSeek Harness: those are coding CLI/harness products; Ruflo is a meta-harness on top of them. Vs Paperclip: Ruflo = execution swarm inside tools; Paperclip = org/budget governance. Extreme ★ velocity — treat discovery heat as noise until the swarm loop proves out on your stack.

• BuzzApache-2.0 self-host Nostr event-log workspace for humans+agents

  Free / OSS · Apache-2.0 · 36.2k GitHub stars

  VerdictApache-2.0 self-host human+agent rooms when a keyed event-log workspace beats another board or messaging bot — expect prototype heat and issue volume.

  CatchVs OpenClaw: messaging channels/employee, not a Nostr event-log collab workspace. Vs Multica: coding-agent issue board. Vs Paperclip: org budgets/Board governance. Org maturity reads prototype; ~3.6k open issues; Block-internal paths and optional Railway one-click are extras, not OSS walls.

• Matt Pocock SkillsMIT composable SKILL.md pack for real engineering — not vibe coding

  Free / OSS · OSS/MIT · 299k GitHub stars

  VerdictMIT real-engineering skill pack when agents need grilling, TDD, and deep-module habits — rank the workflows, not the ★ counter.

  CatchStar counts on viral skill repos are a weak signal (watchers are far below stars here); judge it on substance, not the counter. Vs agent-skills (Addy Osmani `/spec→/ship` lifecycle pack) and Superpowers (session-hijacking brainstorm→plan→TDD methodology): this is Matt’s composable engineering/productivity skill library you pick and adapt, not a Google-flavored lifecycle gate set and not an automatic SDLC hijack. Installing both the Claude plugin and skills.sh copies doubles every skill — pick one path.

• OpenShellApache-2.0 agents/sandbox policy runtime — deny-by-default kernel governance

  Free / OSS · Apache-2.0 · 16.8k GitHub stars

  VerdictApache-2.0 policy runtime when agents need deny-by-default sandbox governance — not E2B/Daytona cloud seats; rank the YAML, not the ★.

  CatchVs E2B / Daytona / CubeSandbox: those are hosted Firecracker sandboxes, freemium stateful env computers, or self-hosted RustVMM/KVM density — OpenShell’s catch is deny-by-default kernel governance + credential injection as policy, not a sandbox SaaS seat. Telemetry defaults on (disableable). ★10,805 in about 60 weeks (~180 a week), so treat the star count with caution; rank on substance, not the counter.

More with this tag (20)

• SkillOpt — MIT skills optimizer — trains skill.md as params for frozen agents
• Magnitude — Apache-2.0 local inference desktop+CLI — consumer profiler for agents
• Cognee — Apache-2.0 AI memory platform — local knowledge graph for agents
• Herdr — Apache-2.0 terminal runtime for multi-agent herds — state, panes, socket API
• Zeron — MIT local control plane for coding agents — optional multi-device sync
• Nasiko — Apache-2.0 developer control plane — discover, route, and trace agent harnesses
• Open-Science (AIPOCH) — Apache-2.0 agents research workbench — brew cask / download free
• iFixAi — Apache-2.0 agents eval/audit — pip install, premium ≠ paywall
• BMAD Method — MIT agent skills method — npm bmad-method, not proprietary
• Headroom — Apache context-compression layer — shrink tool dumps before the model
• Langflow — MIT visual agent/workflow builder — every flow as API + MCP tool
• kagent — Apache Kubernetes-native framework for building AI agents
• OpenDesign — Apache collaborative design-agent workspace — Claude Design alt
• OpenSRE — Apache-2.0 self-host SRE agent — Ollama path, hosted quickstart is not free
• Cloudflare OS — Apache-2.0 AI productivity OS reference design — local trial via pnpm run-local
• eve — Apache-2.0 filesystem-first framework for durable AI agents — from Vercel
• tuios — MIT terminal window manager that shows which of your coding agents is working, waiting or done
• mcp-agent — Apache-2.0 MCP-as-runtime + durable Temporal workflows (LastMile)
• Windmill — AGPLv3 scripts→workflows/UIs/agents; CE free self-host unlimited runs
• Daytona — Freemium agent sandboxes — stateful envs, SDK/API/CLI