82 entries · verified 2026-09-24 · schema agent-tool.schema.json
Agent Tools: 82 verified entries, 82 of which need no credit card, last checked 2026-09-24.
“Free plan available” is the least useful sentence in developer marketing. This page puts the actual number next to each tool: Gemini CLI's 1,000 free requests per day, GitHub Copilot Free's 2,000 completions, Windsurf's unlimited tab completion with a quota-limited agent.
The cheapest credible setup in 2026 is an open-source agent pointed at a free API tier: Cline or Aider with a Gemini or Groq key costs nothing and has no request ceiling beyond the provider's. The commercial editors are included because their free tiers are genuinely usable for evaluation, not because they are the cheapest path.
Frameworks are listed with what they are actually good at rather than a feature matrix — LangGraph for stateful runs, LiteLLM to stack several free tiers behind one endpoint, n8n when the agent needs to touch existing systems.
showing 82 of 82
Local-first desktop agent with self-evolving skills and multi-model support, built as a privacy-preserving alternative to hosted coworking agents.
✅ no credit cardopen source
Mid-sized community project. "Self-evolving skills" means it writes new skills for itself over time — keep an eye on what it accumulates.
Open-source Zapier alternative with AI pieces and MCP support, MIT-licensed rather than fair-code.
✅ no credit cardopen source
Microsoft-originated multi-agent conversation framework: agents that talk to each other until the task is done.
pip install ag2✅ no credit cardopen source
Scanner and self-hosted control plane for AI infrastructure — sweeps agent configs, MCP servers, instruction files, containers, cloud estate and model artefacts.
pip install agent-bom✅ no credit cardopen source
The agent-specific targets — instruction files and MCP configs — are what distinguish it from a generic container scanner. Actively released.
Evaluation framework that auto-generates test tasks for an agent, executes them and scores results across several dimensions.
✅ no credit cardopen source
Auto-generated tasks solve the blank-page problem in agent testing, but they measure what the generator thought to ask — combine with your own regression cases.
High-performance multi-modal agent framework with built-in memory, knowledge and reasoning tools.
pip install agno✅ no credit cardopen source
Terminal pair-programmer that commits every change to git and maps your repo into the model's context.
pip install aider-install && aider-install✅ no credit cardopen source
Single-compose self-hosted AI platform bundling inference, tool use, browser automation, image generation, speech synthesis, transcription and object storage.
docker compose up✅ no credit cardopen source
Small project with a wide attack surface — one compose file exposing inference, storage and code execution. Read the compose file before exposing it beyond localhost.
Local-first continuity layer that lets agents resume interrupted work, track open loops and recall decisions with provenance, exposed over MCP, HTTP and CLI.
uvx alice-memory mcp✅ no credit cardopen source
Provenance — remembering why a decision was made, not just what it was — is the part most memory layers skip. The 79.4% LongMemEval figure is the project's own reported number, not independently reproduced here.
Frontier agent from Sourcegraph with a free-forever plan, then pay-as-you-go with no subscription required.
✅ no credit card
Agentic RAG stack assembled entirely from open-source components, with nine swappable layers so you can replace any piece without touching the rest.
✅ no credit cardopen source
Worth reading as a reference architecture even if you do not adopt it — the swappable-layer structure is a clean answer to RAG stacks that hard-code one vendor.
Apache-2.0 multi-agent orchestration platform built around RAG and vector embeddings, able to route through free provider tiers including NVIDIA's.
✅ no credit cardopen source
Judge it as a multi-agent orchestration stack; the "neural field theory" framing is not an evaluable claim.
Security scanner for MCP servers and AI skill files that scores findings against OWASP's AIVSS rubric and ships an open vulnerability database with a public API.
pip install bawbel-scanner✅ no credit cardopen source
Scoring MCP servers against a published rubric is useful because it makes two servers comparable. AIVSS is still evolving, so treat scores as guidance rather than verdicts.
Multi-dimensional agent evaluation that scores runs across ten dimensions and visualises the spread as radar charts.
✅ no credit cardopen source
Multi-dimensional scoring is more informative than a single pass/fail number. The advertised "17-judge AI tribunal" and deception detection are separate LLM calls judging output — treat their verdicts as signal, not ground truth.
Open-source agentic browser — a Chromium-based alternative to ChatGPT Atlas, Perplexity Comet and Dia, with agents built into the browser rather than bolted on.
✅ no credit cardopen source
The submission pointed at a one-star fork (gHashTag/BrowserOS); this is the upstream project. AGPL-3.0 matters if you plan to redistribute a modified build.
Self-hosted sandbox for AI agents, built as a community alternative to hosted coworking sandboxes.
✅ no credit cardopen source
Young project — verify the isolation boundary yourself before trusting it with untrusted code, since that is precisely the promise being made.
Autonomous coding agent for VS Code, JetBrains and the terminal, with human approval at every step.
✅ no credit cardopen source
The most flexible free setup: point it at Gemini or Groq free tiers and a full agent costs nothing.
Local-first coding agent that reads your repo, writes code and runs commands, with terminal, HTTP and optional desktop interfaces plus fleet collaboration.
npm install @phuetz/code-buddy✅ no credit cardopen source
The Ollama path is the supported free route and is why this is listed. Its optional ChatGPT-subscription mode routes through a consumer account rather than an API, which is outside that provider's terms — use the local path.
Agent-agnostic framework for evaluating and benchmarking coding agents and skills, with support for comparing different agent harnesses head to head.
pip install coder-eval✅ no credit cardopen source
Agent-agnostic is the valuable property: you can A/B your own agent against a reference harness instead of trusting a vendor's self-reported score.
Open-source memory platform that builds a self-hosted knowledge graph from your data, giving agents persistent long-term memory across sessions.
pip install cognee✅ no credit cardopen source
Runs end to end without an OpenAI or Anthropic key by using local extraction and embedding models, which is unusual for knowledge-graph memory. A hosted tier exists for teams that would rather not self-host.
Open-source Copilot alternative for VS Code and JetBrains with customisable model routing.
✅ no credit cardopen source
Role-based multi-agent crews with delegation, sequential and hierarchical process modes.
pip install crewai✅ no credit cardopen source
Charm's glamorous terminal agent with MCP support, LSP awareness and session management.
brew install charmbracelet/tap/crush✅ no credit cardopen source
AI-first code editor with whole-project awareness, tab completion and a multi-file agent.
✅ no credit card
Free tier is an evaluation budget, not a workflow. Pair it with Gemini CLI or Cline for the heavy lifting.
Agentic editor from Cognition — the Windsurf/Cascade lineage, renamed after the 2026 acquisition.
✅ no credit card
windsurf.com now redirects to devin.ai. Tab completion is unlimited on every plan, including free — the agent quota is what runs out.
LLMOps platform for building chat agents, workflows and RAG apps with a visual editor and API output.
✅ no credit cardopen source
Programs language models instead of prompting them — optimisers compile prompts against a metric.
pip install dspy✅ no credit cardopen source
Low-code LLM orchestration with agent flows, tools and embeddable chat widgets.
npx flowise start✅ no credit cardopen source
Coding agent funded by terminal advertising — desktop, CLI, web and chat clients all backed by frontier models at no cost, in exchange for text ads.
✅ no credit card
Marked degraded rather than active because the business model is the product: the vendor sells terminal ad impressions to fund inference, so model access and the ad load can both change without notice. Closed source, and the "unlimited" wording is marketing — throughput is throttled in practice. Do not point anything production-critical at it.
Google's open-source terminal agent with the most generous free quota of any first-party CLI.
npx -y @google/gemini-cli✅ no credit cardopen source
Also ships a free Gemini Code Assist VS Code extension on the same account. The best $0 coding agent in 2026.
Copilot's genuinely free tier inside VS Code, JetBrains and the GitHub web editor.
✅ no credit card
Requires any GitHub account, including free. Best free choice if you already live in VS Code.
Google's agent framework with a built-in dev UI, evaluation harness and A2A protocol support.
pip install google-adk✅ no credit cardopen source
Block's local-first agent that can extend itself by writing its own MCP extensions.
curl -fsSL https://github.com/block/goose/releases/download/stable/download_cli.sh | bash✅ no credit cardopen source
Agentic RAG that runs locally by default, using Ollama-hosted open models for the full pipeline — ingestion, embeddings, retrieval and generation.
pip install haiku.rag✅ no credit cardopen source
Scales from a laptop to a server without changing the stack, and the no-key default is the reason it belongs in this catalogue rather than in a generic RAG list.
Production RAG and agent pipelines from deepset, with strong evaluation and deployment tooling.
pip install haystack-ai✅ no credit cardopen source
Antivirus and runtime protection layer for agent tooling — plugins, skills and MCP servers — that sits between your agent and the code it is about to run.
pip install hol-guard✅ no credit cardopen source
Installing an MCP server means running someone else's code with your credentials in reach; this is one of the few tools treating that as the security boundary it is. Integrates with the major agent CLIs.
Desktop coding workbench where the agent reads the project, runs commands, drives a browser and coordinates subagents — asking before it changes anything.
npm install hyperdesk✅ no credit cardopen source
The confirm-before-changing behaviour is the right default for a tool with this much reach. Young project — check the package's own release notes before trusting it with a large repo.
Open-source VS Code agent with a large free model allocation before you need your own key.
✅ no credit cardopen source
AWS's spec-driven agentic IDE — writes requirements and design docs before it writes code.
✅ no credit card
Local-first agent framework that runs as a VS Code extension plus web and mobile clients, routing to any of hundreds of models across nine provider categories.
pip install kiss-agent-framework✅ no credit cardopen source
The "KISS" in the name is the design brief. Provider counts and third-party agent integrations move quickly in this project, so treat any specific number (661 models, 43 agents) as a snapshot rather than a spec.
Durable-execution runtime for Python agents from the ZenML team: record, replay and improve agent runs in production.
pip install kitaru✅ no credit cardopen source
Submitted as "ZenML/Kitaru"; the canonical repository is zenml-io/kitaru. Built by the team behind ZenML, which is the main reason to trust it at this age.
The widest integration ecosystem in the space — 600+ tools, loaders and vector store connectors.
pip install -U langchain✅ no credit cardopen source
Visual drag-and-drop builder that compiles to LangChain-based flows and exposes each as an API.
uvx langflow✅ no credit cardopen source
Stateful, graph-based agent orchestration with durable execution, human-in-the-loop and time travel.
pip install -U langgraph✅ no credit cardopen source
Choose this over raw LangChain when your agent needs to pause, resume or survive a crash mid-run.
Stateful agents with long-term memory that manage their own context window (formerly MemGPT).
pip install letta✅ no credit cardopen source
One OpenAI-shaped interface over 100+ providers, with a proxy that adds budgets, keys and tracing.
pip install litellm✅ no credit cardopen source
The cleanest way to stack several free tiers behind one endpoint and fail over automatically when one 429s.
Data framework for RAG: ingestion, chunking, indexing and query engines over your own documents.
pip install llama-index✅ no credit cardopen source
Chat with any codebase locally: AST-aware chunking, Ollama embeddings and a Chroma vector store, with no code leaving your machine.
✅ no credit cardopen source
AST-aware chunking beats naive line splitting on code, because a function is the unit you actually want retrieved. Small community project — read before you run.
TypeScript agent framework with workflows, evals, RAG and local dev playground.
npm create mastra@latest✅ no credit cardopen source
Coding-agent CLI with a memory and checkpoint layer: it keeps a queryable record of project decisions, conventions and past mistakes across sessions.
pip install mazu✅ no credit cardopen source
Pair it with any of the free coding models catalogued here. As with every local memory store, the retrieval quality matters more than the storage format.
Drop-in memory layer for agents that extracts and stores salient facts between sessions, with the widest framework support of any memory library.
pip install mem0ai✅ no credit cardopen source
The most-starred memory project by a wide margin. Extraction-based like its benchmarks measure against — it summarises rather than storing verbatim, so recall depends on the extraction step. The hosted platform is a paid product beyond the free allowance.
Local-first memory for agents that stores conversations verbatim in ChromaDB and organises them into wings, halls and rooms instead of letting an LLM decide what is worth keeping.
pip install mempalace && mempalace init✅ no credit cardopen source
Co-created by Milla Jovovich and Ben Sigman.
Often quoted as "100% on benchmarks" — that is the hybrid pipeline with cloud reranking, which the authors themselves flagged as tuned on failing questions. The defensible numbers are 96.6% R@5 raw on LongMemEval with zero API calls and a 98.4% held-out hybrid score.
The project publicly walked back an overstated "30x lossless compression" claim; its experimental AAAK mode actually regresses to 84.2%.
Human-like memory layer for agents with auto-save, auto-recall and a cognitive profile, split across semantic, episodic and procedural memory and exposed as MCP tools.
pip install mengram-ai✅ no credit cardopen source
Splitting memory into semantic, episodic and procedural stores is a more honest model of how agents actually benefit from memory than a single flat vector index. Actively released and includes Claude Code hooks.
Offline coding agent for the terminal that runs entirely on Ollama, positioned as a local alternative to Claude Code, Cursor and Copilot.
✅ no credit cardopen source
Small but actively developed. Output quality tracks whichever local model you point it at, so pair it with a decent coding model rather than a 3B.
Node-based automation with native AI agent nodes, MCP trigger and hundreds of integrations.
npx n8n✅ no credit cardopen source
Running the Docker image on a free-tier VM is the cheapest credible production agent host in this list.
Deliberately tiny personal-agent framework from the HKU Data Intelligence lab that reimplements most of a large agent stack in roughly 4K lines.
✅ no credit cardopen source
Actively developed by a research lab and worth reading end-to-end precisely because it is small. The "90% of OpenClaw's capabilities" figure is the project's own framing, not a measured comparison.
Open-source Chrome extension that runs multi-agent web workflows — extract, compare and fill forms — inside your own browser, on your own keys.
✅ no credit cardopen source
The pitch is "open-source alternative to OpenAI Operator", and its real advantage is BYOK: pair it with Ollama for a fully local browser agent, or with a free Groq/Gemini key. Chrome-family browsers only.
Self-hosted orchestration platform where agents get personas, skills and tools, with sub-agent delegation, RAG and semantic memory.
✅ no credit cardopen source
Early-stage (tens of stars, commits this month). Useful if you want a self-hosted multi-agent UI today; audit it before pointing it at anything sensitive.
Community continuation of Daytona: elastic sandbox infrastructure for running AI-generated code and agent workflows in isolation.
✅ no credit cardopen source
Forked from Daytona's last AGPL-3.0 release after that project changed direction, so the licence is what makes the continuation legitimate — and it means you inherit the maintenance burden rather than a vendor's.
Desktop agent for Windows and macOS that installs Claude Code and MCP tooling in one click and runs the agent in a sandbox.
✅ no credit cardopen source
Sandbox isolation is the feature that matters here — the whole point of this class of tool is letting an agent act on your machine, and the sandbox is what bounds the blast radius.
Small, provider-agnostic agent SDK: handoffs, guardrails, sessions and tracing in a few hundred lines.
pip install openai-agents✅ no credit cardopen source
Despite the name, it is vendor-neutral — point it at Gemini or Groq and it works unchanged.
Open-source terminal coding agent with a client/server architecture and provider-agnostic model routing.
curl -fsSL https://opencode.ai/install | bash✅ no credit cardopen source
Open-source software engineering agent (formerly OpenDevin) that runs in a sandboxed container.
docker run -it --rm -p 3000:3000 docker.all-hands.dev/all-hands-ai/openhands:latest✅ no credit cardopen source
Self-hosted benchmarking harness for computer-use agents acting through MCP, letting you measure tool-using agents reproducibly.
✅ no credit cardopen source
Computer-use evaluation is the least standardised corner of agent testing, so any harness that makes results reproducible is worth knowing about. Early-stage project.
One-command monitoring and guarding for agents and MCP servers, aimed at catching what an agent does rather than what it was asked to do.
npm install panguard✅ no credit cardopen source
Deliberately low-friction — one command and it starts watching, which is the right trade for teams that will not stand up a full security stack.
Desktop agent for macOS and Windows with free models bundled in, aimed at office files, web search and scheduled automations rather than code.
✅ no credit cardopen source
The no-key, no-terminal framing makes it a genuine on-ramp for non-developers. Bundled free models change without notice, since they depend on upstream providers.
Minimal, inspectable coding agent that persists project facts to a plain .pi/memory.md file instead of a hidden database, with self-review and autonomous tool use.
pip install pi-coding-agent✅ no credit cardopen source
Storing memory as readable Markdown you can edit and review is the whole appeal — no opaque vector store to debug when recall goes wrong.
Portable, agent-first knowledge base where one directory holds the corpus, retrieved with SQLite FTS5 lexical search plus local embeddings and local reranking.
pip install pinakes✅ no credit cardopen source
Combining BM25 with embeddings is more robust than either alone — lexical search catches exact identifiers that pure vector search misses, and re-indexing costs nothing.
Self-hosted personal agent that lives in your messaging apps — WhatsApp, Telegram, Slack, Signal and email — with a large native tool set plus everything reachable over MCP.
pip install pincer-agent && pincer init✅ no credit cardopen source
Skill signing and audit logs are the differentiators: an agent wired into your WhatsApp can take real actions on your behalf, so its security posture matters more than a coding agent's.
On-device RAG API built on a ~300M-parameter model small enough to run on a Raspberry Pi 5, with no GPU and no external API calls.
✅ no credit cardopen source
Genuinely edge-sized, which is rare for anything labelled RAG. A 300M model is for retrieval and grounded short answers, not general reasoning.
Type-safe agent framework from the Pydantic team — validated structured outputs and dependency injection.
pip install pydantic-ai✅ no credit cardopen source
Qwen's fork of Gemini CLI wired to Qwen3-Coder, with a large free daily request allowance via OAuth.
npm install -g @qwen-code/qwen-code✅ no credit cardopen source
Strong on large refactors thanks to Qwen3-Coder's context window. The 2,000/day figure is subject to change.
Personal assistant agent that runs fully locally against its own bundled QwenPaw-Flash models, so it needs no API key and no cloud account at all.
pip install qwenpaw✅ no credit cardopen source
Genuinely key-free out of the box rather than BYOK, which is rare. Connectors skew to Chinese platforms (DingTalk, Lark, WeChat, QQ) alongside Discord, Telegram and iMessage. A 2B-9B local model is a capable assistant, not a frontier-reasoning substitute.
Self-hosted workflow engine and MCP toolkit that records agent runs so you can replay them offline, fork a run, diff two runs, or freeze one into a regression test.
pip install readyagentsdev✅ no credit cardopen source
The record/replay/diff loop is the genuinely unusual part — most agent frameworks still have no first-class way to turn a production run into a regression test.
Cline fork with custom modes (architect, debugger, tester) and an orchestrator mode for multi-step work.
✅ no credit cardopen source
Microsoft's enterprise agent SDK for Python, .NET and Java with planners and process framework.
pip install semantic-kernel✅ no credit cardopen source
Hugging Face's minimal agent library — code-writing agents in under 1,000 lines of core logic.
pip install smolagents✅ no credit cardopen source
Self-hostable data plane for always-on agents: ingest events from Prometheus, OpenTelemetry and webhooks, then let an MCP server act on them.
pip install tares✅ no credit cardopen source
Targets the gap between "an agent that answers" and one that reacts to what your infrastructure is actually doing.
ByteDance's AI IDE with a free tier built around frontier models and an agent mode.
✅ no credit card
Review the data-retention policy before using it on proprietary code.
ByteDance's multimodal GUI agent stack that drives a real desktop — clicking, typing and navigating native apps — from screenshots and instructions.
✅ no credit cardopen source
One of the few computer-use stacks with real industrial backing and a permissive licence. A GUI agent that can click anything on your screen is a large security decision — run it in a VM or on a spare machine.
The standard TypeScript toolkit for streaming UI plus agents, with unified provider interfaces.
npm install ai @ai-sdk/openai✅ no credit cardopen source
Rust-native, GPU-accelerated editor with a free built-in AI tier and bring-your-own-key support.
✅ no credit cardopen source
There is no single best answer, because the honest options split three ways: open-source agents you run and point at your own model key, vendor CLIs with a genuinely free tier, and IDE-integrated agents bundled with an editor. This catalogue lists 82 such tools and records what each one's free access actually costs you.
82 are listed here as of 2026-09-24, spanning 0 vendor-maintained tools and 75 open-source projects. They range from terminal agents to orchestration frameworks, and each entry states whether a card is required and what the free limit is.
Mostly they are free software, not free compute. An open-source agent costs nothing to download but still needs a model, so the real cost is the API key you give it — which is why this catalogue pairs agents with free LLM API providers that need no card.
An agent completes tasks by itself and reports back. An IDE-integrated assistant works inside your editor while you type. A framework is a library you write code against to build your own agent. All three are catalogued here, tagged by category, because people searching for one often need another.