ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
ToolTypeWhen limits resetHow much you get freeShort summaryModels & featuresCons / downsidesInstall commands (Windows / macOS / Linux-Fedora)StatusPrimary source
2
Google Antigravity (agy CLI + IDE + 2.0 desktop app)CLI + IDE + desktop appWEEKLY. Free quota refreshes weekly; unused allowance does not roll over. (Paid plans refresh every 5 hours up to a weekly cap.)$0 Individual plan, Google sign-in, no card. 'Basic weekly rate limits' on agent models (no numbers published); Tab completions and Command requests are unmetered. Gemini and third-party models have separate weekly pools.Google's agent-first dev platform: IDE, Antigravity 2.0 desktop app, agy terminal CLI and editor extensions. It replaced Gemini CLI for free and individual accounts on 2026-06-18.Free column: Gemini 3.8 Flash; plus Claude Sonnet 4.6 / Opus 4.6 and GPT-OSS-120B until they are removed on 2026-11-18 (some sources said 11-02).
Features: plan mode, subagents, hooks, skills, MCP (mcp_config.json), /schedule cron tasks, browser agent, AGENTS.md / GEMINI.md, 'agy plugin import gemini', extensions for VS Code, JetBrains, Zed, Xcode.
Quota is unpublished and small; agents burn it fast.
After 2026-11-18 the free tier is reportedly Gemini Flash only.
May train on your data unless you opt out.
HTTP 429 / quota-lockout bugs reported; past safety incidents (Dec-2025 drive wipe in Turbo mode; a prompt-injection flaw, patched).
Google banned accounts that connected the third-party agent OpenClaw.
No bring-your-own-key route for extra limits. Closed source.
Windows: irm https://antigravity.google/cli/install.ps1 | iex
(CMD: curl -fsSL https://antigravity.google/cli/install.cmd -o install.cmd && install.cmd && del install.cmd)
macOS: curl -fsSL https://antigravity.google/cli/install.sh | bash (or: brew install --cask antigravity-cli)
Linux (Fedora): curl -fsSL https://antigravity.google/cli/install.sh | bash
Then run: agy
Active - changing soonantigravity.google/pricing; github.com/google-gemini/gemini-cli/discussions/28017
3
OpenCode (with Zen free models)CLI/TUI + desktop app + IDE extensionNOT PUBLISHED. Free models rotate and are each 'available for a limited time'. Only LongCat 2.5 has a stated cap: 100 requests/day.$0, no sign-in needed for Zen free models. Usage amount not published. Free models work only inside OpenCode (other clients get 403). Any other provider via bring-your-own-key.Open-source terminal coding agent with a built-in gateway (Zen) that serves a rotating set of free hosted models.Free as of 2026-11-12: GLM-5.3-Flash, DeepSeek V4.1 Flash, MiniMax M3.1 Flash, MiMo-V2.6-Flash, Nemotron 3 Ultra / 3.5 Super VL / 3.5 Lightning, Big Pickle, Beam, Flint 40B, Inkling 2 Small, LongCat 2.5.
Features: TUI, desktop app, MCP, any provider via BYOK, plan/build agents.
Free list churns every few weeks (models added/dropped constantly).
Free-period prompts may be used to train models; some models are China-hosted or exclude Russia.
No published quota; weaker than frontier models.
Windows works best under WSL.
Windows: scoop install opencode | choco install opencode | npm install -g opencode-ai (WSL: curl -fsSL https://opencode.ai/install | bash)
macOS: brew install anomalyco/tap/opencode (or: curl -fsSL https://opencode.ai/install | bash)
Linux (Fedora): curl -fsSL https://opencode.ai/install | bash (or: npm install -g opencode-ai; the desktop app also ships an .rpm)
Then run: opencode
Activeopencode.ai/docs/zen; mvalentsev.github.io/awesome-free-ai-coding/providers/opencode
4
Kilo Code (CLI, VS Code, JetBrains)CLI + IDE extensionsROLLING HOURLY per IP on the keyless gateway (200 requests/hour). Free-model quotas have no published reset.Free: 'Auto Free' router plus free models via the Kilo Gateway, including a no-account keyless OpenAI-compatible lane (200 requests/hour per IP). BYOK and local models cost Kilo nothing.Open-source agent (the CLI is a fork of OpenCode) with a gateway to 500+ models. The only entry here with a no-account free lane.Free (rotating): Nemotron 3.5 Super VL, Nemotron 3 Ultra, Nemotron 3.5 Lightning, North Mini Code, dots-3-note, Flint v1.0, plus openrouter/free and kilo-auto/free routers that pick a model for you.
Features: architect/code/debug modes, MCP, sandboxed auto mode, parallel agents, IDE autocomplete.
The router picks the model, not you.
Free prompts may be used for training; lineup rotates.
Rate-limited: fine for light work, not marathon agent sessions.
Anything beyond the free lane uses paid credits.
Windows: npm install -g @kilocode/cli (or download kilo-windows-x64.zip from GitHub Releases)
macOS: npm install -g @kilocode/cli (or kilo-darwin-arm64.zip)
Linux (Fedora): npm install -g @kilocode/cli (Node 18+; or kilo-linux-x64.tar.gz)
Then run: kilo
Activekilo.ai; mvalentsev.github.io/awesome-free-ai-coding/providers/kilo-code
5
GitHub Copilot Free (+ Copilot CLI)IDE extension + CLIMONTHLY.$0, no card. 2,000 completions/month plus a limited allowance of GitHub AI Credits for chat, agent mode and Copilot CLI (amount unpublished; you cannot buy more on Free).GitHub's assistant. The Free plan includes IDE agent mode and the Copilot CLI. Sign in with GitHub.Auto model selection only (entry-level models such as GPT-5 mini and Claude Haiku 4.5, per GitHub).
Features: agent mode in VS Code / Visual Studio / JetBrains / Eclipse / Xcode, Copilot CLI, chat on github.com and mobile, MCP.
No model picker; credit allowance unpublished (some older GitHub FAQ text still says 50 chat requests).
No cloud coding agent and no code review on Free.
Free-tier prompts may be used for training unless you opt out.
Individuals only: not available if your org provides Copilot.
Windows: winget install GitHub.Copilot
macOS: brew install copilot-cli (or: npm install -g @github/copilot)
Linux (Fedora): curl -fsSL https://gh.io/copilot-install | bash (or: npm install -g @github/copilot, Node 22+; or Homebrew)
Then run: copilot
Activegithub.com/features/copilot/plans; docs.github.com/copilot/how-tos/copilot-cli/install-copilot-cli
6
Freebuff (Codebuff)CLI + desktop + web + cloudDAILY. Hours refill at midnight local time; unused hours do not carry over.Ad-funded, no card. 30 hours/day in 24 listed countries (US, CA, UK, AU, NZ, IE, Nordics, most of W. Europe, S. Korea); 5 hours/day elsewhere; 4 hours/day on a VPN. Balance is counted in hours, not tokens.Ad-supported coding agent from the Codebuff team. Text ads show during sessions in exchange for free access to the included models.Default: DeepSeek V4.1 Flash. Also GLM 5.3 Flash, MiMo V2.6 Flash, Solar Mini 4 / Pro 4. GPT-6 Luna, GPT-6.1 Terra and MiMo 2.6 Pro only at 'full access'; GPT-6.1 Sol and Gemini 4 Argon need a paid plan.
Features: specialist subagents (file picker, code reviewer), web research, real-browser testing, parallel agents (Desktop), cloud sandbox for GitHub repos, web app builder.
Text ads in your sessions.
Collects prompts, code, repo data and agent traces (may train).
Allowance depends on country / VPN; 'promotional prices are temporary'.
Older accounts keep legacy models only until they are retired.
Windows: npm install -g freebuff (Desktop installer at freebuff.com)
macOS: npm install -g freebuff
Linux (Fedora): npm install -g freebuff (needs Node.js)
Then run: freebuff
Activegithub.com/CodebuffAI/freebuff; mvalentsev.github.io/awesome-free-ai-coding/providers/freebuff
7
Cline (CLI, VS Code, JetBrains, Desktop)CLI + IDE extensions + desktopDAILY, per model (the in-app limit card shows the exact reset time).Rotating FREE models once you sign in to a Cline account (Google / GitHub / email, no card). Daily cap per model, amount unpublished. Free quotas do not cover the Cline API.Open-source (Apache-2.0) autonomous coding agent. Free models are promotions inside the Cline provider, in the extension, CLI and desktop app.Free now: Nemotron 3.5 Super VL, Beam 1.0, Nemotron 3.5 Lightning, Solar Mini 4 (rotates).
Features: plan/act, MCP marketplace, checkpoints, rules/skills, scheduled agents (cline schedule), BYOK for any provider, local models.
Free list rotates; daily caps unpublished.
Free usage may be used to improve models.
A May-2026 guide called the CLI mac/Linux-first (Windows CLI 'on the roadmap'); the VS Code extension works everywhere.
Other models are pay-as-you-go.
Windows: npm install -g cline (or VS Code: code --install-extension saoudrizwan.claude-dev)
macOS: npm install -g cline
Linux (Fedora): npm install -g cline (Node 20+)
Then run: cline auth (choose 'Sign in with Cline')
Activedocs.cline.bot/getting-started/quick-start; mvalentsev.github.io/awesome-free-ai-coding/providers/cline
8
Kiro (AWS) - IDE, CLI, webIDE + CLI + webMONTHLY. Credits do not roll over.50 credits/month on the perpetual Free plan, no card (sign in with GitHub, Google or AWS Builder ID).AWS's spec-driven agentic IDE and CLI, the successor to Amazon Q Developer.Free-column models: Claude Sonnet 4 / 4.5, DeepSeek V3.2, GLM-5, MiniMax M2.1 / M2.5, Qwen3-Coder-Next.
Features: spec-driven workflow (requirements > design > tasks), hooks, steering files, MCP, headless CLI for CI.
50 credits go fast on big specs.
Newer Sonnet / Opus / GPT tiers are outside Free.
Free requests are served from the US (GovCloud excluded); may train unless you opt out.
Linux IDE needs glibc 2.39+ (Fedora 40+ OK).
Windows: irm 'https://cli.kiro.dev/install.ps1' | iex
macOS: curl -fsSL https://cli.kiro.dev/install | bash
Linux (Fedora): curl -fsSL https://cli.kiro.dev/install | bash (IDE: .rpm / .deb / AppImage from kiro.dev/downloads)
Then run: kiro-cli
Activekiro.dev/docs/getting-started/installation; mvalentsev.github.io/awesome-free-ai-coding/providers/kiro
9
Google Jules (+ Jules Tools CLI)Async cloud agent (web + CLI)ROLLING 24 HOURS.15 tasks per rolling 24 h, 3 running at once. No card. Limits follow your Google AI plan.Google's asynchronous cloud agent: clones your GitHub repo into a Google Cloud VM, plans, edits, runs tests and opens a pull request.Gemini 3 Flash is the base model on Free (older doc tables still list Gemini 2.5 Pro).
Features: editable plan before it runs, GitHub label triggers, Jules Tools CLI, public API, reusable environment snapshots.
GitHub repositories only.
Fire-and-forget, not interactive; Free uses the smaller model.
Google may adjust limits at any time.
Windows: npm install -g @google/jules (web app needs no install: jules.google)
macOS: npm install -g @google/jules
Linux (Fedora): npm install -g @google/jules (needs Node.js)
Then run: jules login
Activejules.google/docs/usage-limits; mvalentsev.github.io/awesome-free-ai-coding/providers/google-jules
10
Cursor (Free) + Cursor CLIIDE + CLIUNCLEAR. Cursor's pricing FAQ describes weekly refreshes; its docs do not confirm a Free reset. Check your account.Limited Agent, Chat and Tab usage on Cursor's own models. Numbers unpublished; third-party models not included.AI-native VS Code fork plus a terminal agent (the 'agent' command).Free models listed by the verified registry: Composer 2.5 (first-party) and Grok 4.5 / 4.6 / 4.7.
Features: Agent / Plan / Ask modes (Shift+Tab), headless mode for CI, rules, MCP.
No published numbers; reset behaviour ambiguous.
No third-party frontier models on Free.
May train unless Privacy Mode is on.
Old '2,000 tab + 50 premium requests' figures are outdated.
Windows: irm 'https://cursor.com/install?win32=true' | iex
macOS: curl https://cursor.com/install -fsS | bash
Linux (Fedora): curl https://cursor.com/install -fsS | bash (IDE: download from cursor.com)
Then run: agent
Conflicting sourcescursor.com/docs/cli/installation; mvalentsev.github.io/awesome-free-ai-coding/providers/cursor-hobby
11
Devin Desktop (formerly Windsurf) + Devin CLIDesktop app + CLIDAILY + WEEKLY token budgets for metered models. Free models do not draw from the quota.Free plan: unmetered Tab completions and inline edits; free models that do not consume quota; metered models limited by daily and weekly budgets (amounts unpublished).Cognition's agent app. Windsurf was renamed Devin Desktop on 2026-06-02; the Devin CLI is the terminal side.Free models: SWE-1.6 and SWE-2-Flash-Medium (free until 2026-12-04).
Features: Agent Command Center, Devin Local agent, plan mode, MCP, rules, skills.
I could not confirm the CLI draws from the Free plan.
Free-model promo ends 2026-12-04; Cascade agent removed Sept 2026.
Quotas unpublished; may train on prompts.
Old '25 Cascade credits/month' info is outdated.
Windows: Desktop installer from devin.ai/download; CLI runs under WSL: curl -fsSL https://cli.devin.ai/install.sh | bash
macOS: Desktop from devin.ai/download; CLI: curl -fsSL https://cli.devin.ai/install.sh | bash
Linux (Fedora): Desktop tarball from devin.ai/download (no official .rpm found); CLI: curl -fsSL https://cli.devin.ai/install.sh | bash
Then run: devin
Conflicting sourcesdocs.devin.ai/desktop/devin-desktop-faq; mvalentsev.github.io/awesome-free-ai-coding/providers/windsurf
12
Hermes Agent + Nous Portal (Free plan)Self-hosted agent (CLI + desktop)NOT PUBLISHED.Free Nous Portal plan: rotating ':free' models, no card. Amount and rate caps unpublished.Nous Research's open-source (MIT) self-improving agent. It runs as a process on your machine and can use Nous Portal for hosted models.Free as of 2026-11-12: Beam 1.0, Solar Mini 4, Step 3.7 Flash. Past promos: Nemotron 3 Ultra (June), MiMo V2 Pro.
Features: self-written skills and memory, messaging gateways (Telegram, Discord, Slack, WhatsApp, email), cron scheduler, desktop app.
General-purpose agent, not coding-first.
Free models are short promos and rotate; quotas unpublished.
Portal free plan may train unless you opt out.
Windows: iex (irm https://hermes-agent.nousresearch.com/install.ps1)
macOS: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
Linux (Fedora): curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
Then run: hermes setup (pick Nous Portal > Free plan)
Activehermes-agent.nousresearch.com; mvalentsev.github.io/awesome-free-ai-coding/providers/nous-portal
13
Mistral Vibe CLICLIMONTHLY (billing period), shared with Mistral Studio and the API.Free mode (default for new accounts, no card): a monthly allowance shared across Studio, API and Vibe. Amount not published.Mistral's open-source (Apache-2.0) terminal coding agent. Also runs fully offline with local models.Mistral-hosted models (Devstral 2 family; Medium 3.5 became the default in May 2026 - which ones Free covers is not published) or any local / OpenAI-compatible model.
Features: remote cloud agents, VS Code extension, ACP (vibe-acp), programmatic mode.
Allowance unpublished; usage can stop at the cap until the next period.
API calls may be used to improve Mistral unless you switch off 'Anonymous improvement data'.
Windows: uv tool install mistral-vibe (or: pip install mistral-vibe; Python 3.12+)
macOS: curl -LsSf https://mistral.ai/vibe/install.sh | bash
Linux (Fedora): curl -LsSf https://mistral.ai/vibe/install.sh | bash
Then run: vibe
Activedocs.mistral.ai/vibe/code/cli/install-setup; mvalentsev.github.io/awesome-free-ai-coding/providers/mistral-free
14
Crush (Charm) + Hyper free creditsCLIMONTHLY.100 Hypercredits/month on Charm Hyper's free tier. 1 Hypercredit is about 5 cents of tokens, so roughly $5/month of hosted usage.Charm's polished terminal agent (Bubble Tea TUI, LSP-aware). Hyper is Charm's hosted inference, in beta.Hyper catalog models (zero data retention, no training); BYOK for almost any provider; local models via Ollama.
Features: LSP context, MCP (stdio / http / sse), sessions, permission prompts, 'crush login'.
Only about $5/month of hosted usage.
Hyper is a beta; sign-up may need approval.
FSL-1.1-MIT licence (source-available, not OSI open source).
Windows: winget install charmbracelet.crush (or: scoop bucket add charm https://github.com/charmbracelet/scoop-bucket.git; scoop install crush)
macOS: brew install charmbracelet/tap/crush
Linux (Fedora): add repo, then: sudo yum install crush (repo file /etc/yum.repos.d/charm.repo: baseurl=https://repo.charm.sh/yum/ gpgkey=https://repo.charm.sh/yum/gpg.key; or: npm install -g @charmland/crush)
Then run: crush (crush login hyper)
Activehyper.charm.land/faq; github.com/charmbracelet/crush
15
OpenAI Codex (ChatGPT Free)Desktop app (CLI from Plus)NOT PUBLISHED for Free. (Paid plans: rolling 5-hour window plus a weekly cap.)Limited Codex access on a $0 ChatGPT account, 'subject to rollout'. No number published. Codex Cloud not included.Codex rides your ChatGPT plan. On Free you get the smallest model in the desktop app.GPT-6 Luna at standard speed in the Codex desktop app.
Paid plans add the CLI, IDE extension, web and iOS clients and bigger models.
Per OpenAI's plan cards Free is desktop-app only; the CLI starts at Plus (some third-party guides disagree - try 'codex login' on a Free account).
No published allowance; rollout-dependent; no Codex Cloud.
API-key use is pay-per-token.
Windows: Codex desktop app via developers.openai.com/codex | CLI (Plus+): npm install -g @openai/codex
macOS: Codex desktop app (same page) | CLI: brew install --cask codex or npm install -g @openai/codex
Linux (Fedora): CLI only: npm install -g @openai/codex (or the musl binary from GitHub Releases)
Then run: codex
Conflicting sourceshelp.openai.com/en/articles/11369540; developers.openai.com/codex/pricing
16
Autohand CodeCLI + VS Code / JetBrains + desktopROLLING 5 HOURS + MONTHLY.15 credits per rolling 5 h; 100 credits/month; 10 requests/min. One credit = one successful hosted response. BYOK and local models use no credits.Open-source agent whose Free plan runs Autohand's own hosted model.Autohand's own latency-first model ('fantail'); BYOK with 50+ providers.
Features: 50+ slash commands, plan mode, agent teams, MCP, skills (Codex / Claude formats).
Free needs a linked GitHub account at least 30 days old, an invite / grant, or a card on file (not charged).
About 100 responses a month: light use only; one proprietary model.
Windows: irm https://autohand.ai/install.ps1 | iex (PowerShell installer; check docs.autohand.ai)
macOS: curl -fsSL https://autohand.ai/install.sh | sh (or: brew tap autohandai/code && brew install autohand-code)
Linux (Fedora): curl -fsSL https://autohand.ai/install.sh | sh (or: npm i -g autohand-cli, Node 22+)
Then run: autohand
Activedocs.autohand.ai/working-with-autohand-code/cli; mvalentsev.github.io/awesome-free-ai-coding/providers/autohand-code
17
JetBrains AI Free (Junie agent + Junie CLI)IDE plugin + CLIEVERY 30 DAYS from first use.3 AI credits (about $3 of cloud models) per 30 days for chat and the Junie agent; unlimited local code completion (Mellum). Requires an eligible JetBrains IDE licence.JetBrains' agent (Junie) inside JetBrains IDEs, plus a standalone CLI that is free with your own keys.Cloud models via credits; Junie CLI also supports BYOK (OpenAI, Anthropic, Google, xAI, OpenRouter, GitHub Copilot).
Features: test-verification loop, /plan, /review, /mcp, GitHub Action.
$3/month disappears in one or two tasks.
IntelliJ without Ultimate, PyCharm without Pro and Android Studio are excluded.
Not available in mainland China / provider-restricted regions; may train unless you opt out.
Windows: powershell -NoProfile -ExecutionPolicy Bypass -Command "iex (irm 'https://junie.jetbrains.com/install.ps1')"
macOS: brew tap jetbrains-junie/junie && brew install junie
Linux (Fedora): curl -fsSL https://junie.jetbrains.com/install.sh | bash (or: npm install -g @jetbrains/junie)
Then run: junie
Activegithub.com/JetBrains/junie; mvalentsev.github.io/awesome-free-ai-coding/providers/jetbrains-ai
18
Qoder (IDE + CLI)IDE + CLI + JetBrains pluginDAILY claim (100 credits/day, each claim valid 30 days) plus a one-time 2-week trial.No card. Pro trial: 300 credits once (2 weeks, unmetered completions). Then claim 100 promo credits/day. After that 'basic models' under daily and monthly limits.Agentic IDE / CLI / JetBrains plugin with a promo-driven free allowance.Built-in models billed in credits (model IDs and eligibility must be confirmed in the app); custom models via BYOK (for example DeepSeek).
Features: IDE, qodercli TUI, JetBrains plugin.
Promo-driven and can change; model names not published.
Separate product from Qoder CN (accounts not interchangeable).
Windows ARM64 not supported.
Windows: irm https://qoder.com/install.ps1 | iex
macOS: curl -fsSL https://qoder.com/install | bash
Linux (Fedora): curl -fsSL https://qoder.com/install | bash (or: npm install -g @qoder-ai/qodercli)
Then run: qodercli
Promo / time-limiteddocs.qoder.com/cli/installation; mvalentsev.github.io/awesome-free-ai-coding/providers/qoder
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100