Freebird AI — AI Coding Assistant for VS Code

No setup. No throttling. 20 free AI edits/day.

AI coding assistant · Copilot alternative · Cursor alternative · multi-file AI edits · AI refactoring · codebase search · Claude Haiku · Gemini Flash · Kimi K3 · custom LLM providers (OpenRouter, Together, self-hosted) · prompt templates · Ollama · BYOK · local AI · privacy-first

Run full codebase security audits, architecture mapping, and technical debt analysis with Agent mode — try free for 7 days, no card needed. Pro is $6/month for unlimited multi-file editing, plus sharing a selection with a colleague without giving them your whole codebase — less than a third of Cursor's price, more capability out of the box. Team is $25/month for up to 5 seats.

Install Freebird AI and start coding in seconds — no API keys, no throttling, no configuration. You get 20 free advanced AI edits per day powered by Gemini Flash Lite, plus free BYOK and unlimited local AI when you want full privacy.

Copilot throttled? Cursor too expensive? GitHub limits hit? Freebird never blocks you — it picks up where other tools stop.

Upgrade to Pro — $6 USD/month → · or start a free 7-day Pro trial (no card needed) right from the chat panel

⭐ If Freebird saves you time, a GitHub star helps others find it — thank you!

📅 Heads up: the free daily cloud limit is changing on Monday, September 14. To keep Freebird fast and reliable as usage grows, the free tier's daily cloud limit is moving from 20 to 10 advanced edits/day, effective Monday, September 14. Everything else on Free — unlimited chat, unlimited tab completions on Ollama/BYOK, project rules — is unchanged. If you're already relying on the higher limit day to day, Pro's $6/month unlocks unlimited cloud edits plus full Agent mode (multi-file edits, terminal commands, one-click checkpoints) — or try it free for 7 days, no card needed, right from the chat panel.

Why Freebird

Safe by Design, Not Just "Trust the AI"

Letting an agent touch your codebase should come with proof, not a promise. Every Pro agent turn follows the same rule: nothing changes without your approval, and everything can be undone.

• Approve every edit before it happens — full diff shown, nothing writes silently.
• Checkpoint on every turn — one click restores every file the agent touched back to its exact state before that turn, even several turns later.
• Project-aware, not file-blind — semantic codebase search finds relevant code by meaning ("find the payment retry logic"), not just filename or exact-text matching like a plain grep. It reads your actual workspace tree and fetches the files it needs before proposing a change, and can check git status mid-turn when relevant.
• Terminal commands shown before they run — you see the command, not just the result.

Multi-file edits are only as trustworthy as the undo button behind them — Freebird's is one click, every time.

See It in Action

Multi-file agent edit with Approve / Reject (Pro)

Ask Freebird to update your products page, add images to cards, or refactor across files — it shows a full diff and waits for your approval before changing anything.

Agentic page editing across HTML and CSS (Pro)

Freebird reads your existing code structure, understands the context, and makes targeted edits across files in one agent run.

Codebase Cartographer, run on Freebird's own source (Pro)

We ran our own Codebase Cartographer prompt template against this repo — real output, not staged. It mapped the architecture, traced data flow end to end, and flagged a real, accurate gap in our own auth rollout, all cited to exact files. Below is the Mermaid.js dependency diagram it produced, rendered live in the chat panel.

Free vs Pro

Free tier responses are tagged with the model that answered right in the chat panel — you always know whether you're on the lite model or not, not left to guess.

Team

$25/month for up to 5 seats ($5/seat — cheaper per-seat than an individual Pro subscription) — every Pro feature, one shared subscription. The purchaser becomes the owner and manages teammates from inside the extension with Freebird: Manage Team Seats (add, remove, list — no web dashboard). Contact us to set up a Team plan.

Template Library

33 expert-crafted prompt templates on top of the 3 free ones — each built the same way (a role, a prioritized checklist, "cite the exact file and line," no theoretical padding) and covering real, recurring dev work: migration & modernization (framework migrations, dependency upgrades, state management, CSS), review & quality (senior-engineer-style review, DB migration safety, performance, dead code), compliance & accessibility (WCAG, license compliance, PII, i18n), onboarding & docs (PR descriptions, changelogs, API docs), infrastructure & DevOps (CI/CD, Dockerfiles, IaC), team & process, framework specialists (React, SQL, Python typing, GraphQL), release & ops, and testing deep-dives.

Included free if you're on Pro, Enterprise, or Team — no separate purchase. Otherwise it's a low-cost standalone subscription, cheaper than a full Pro upgrade, for anyone who wants the template library without everything else Pro includes. Browse the whole catalog (locked items are visible, not hidden) from the 📚 icon in the chat panel's top bar, the "Browse prompt templates" welcome-screen prompt, or Freebird: Use Prompt Template in the command palette. Already have a key? Freebird: Activate Template Library License.

What Freebird Replaces

• GitHub Copilot — when you hit your monthly speed limit
• Cursor Composer — multi-file agent edits, without migrating from VS Code
• Claude Code — the same bring-your-own-key workflow, free, without leaving VS Code
• Local coding agents — Ollama integration built in, unlimited and private
• Cloud-forbidden environments — enterprise policies that ban cloud AI, NDA-bound freelance work — Pro's Agent mode runs on Ollama with zero calls ever leaving the machine

Features

Works Immediately — No Setup Required

Install and start coding. Your first 20 advanced edits per day are powered by Gemini Flash Lite — no API key, no Ollama, nothing to configure.

20 Free Advanced Edits Every Day

Unlimited chat, plus 20 cloud-powered inline edits and AI commits a day. Resets daily, no card required. Multi-file agent edits and terminal commands are Pro (see below).

Hit Your Daily Cap? You've Got Options

If you've configured Ollama as your backend, Freebird falls back to it automatically when the cloud quota runs out. Otherwise you'll see a clear "free edits used up" prompt with a no-card 7-day Pro trial one click away — or switch to BYOK (free, unlimited) or Ollama (free, unlimited) anytime from Freebird: Configure AI Backend.

Multi-File Agent Edits with Approve / Reject (Pro)

Freebird reads your codebase, fetches relevant files, and makes targeted edits across multiple paths. Every write shows an Approve / Reject card — nothing changes silently.

Checkpoints — Undo an Entire Agent Turn (Pro)

Every Pro agent turn that creates, edits, copies, or downloads a file gets a "Checkpoint saved" card with a one-click Restore button, reverting every file that turn touched back to its state before the turn. Covers file changes; a turn that also ran a shell command or git push is clearly labeled as not revertible rather than pretending to cover it.

Web Context — Read Any Page the Agent Needs (Pro)

The agent can fetch a webpage's readable content directly — documentation, an article, a URL you paste in — instead of only searching your own codebase. Requests to private/internal addresses (localhost, LAN ranges, cloud metadata endpoints) are blocked, and fetched content is explicitly marked as untrusted reference material before it ever reaches the model, so a malicious page can't pass itself off as an instruction.

Inline Edit — Cursor-style

Select any code, press Ctrl+Alt+K, type an instruction, and the selection is rewritten in place.

AI Tab Completion

Ghost-text suggestions as you type, on every tier — accept with Tab. Runs through whatever backend you've configured, same as the rest of Freebird: instant and unlimited on Ollama, otherwise it's a cloud-powered edit like inline edit and AI commit.

Bring Your Own Keys — Unthrottled, free

Plug in your own Anthropic Claude, OpenAI, DeepSeek, Qwen, or Kimi K3 API key — or point Freebird at any OpenAI-compatible Custom Provider (OpenRouter, Together, self-hosted, etc.). Direct-to-LLM speed, total data privacy, no middleman quotas — free for everyone, since the calls never touch Freebird's servers. DeepSeek in particular is worth a look even if cost isn't a concern: it scores higher than GPT-4o on coding benchmarks and runs about $0.20/million tokens.

Full Agent Mode With Zero Cloud Calls (Pro)

Agent mode isn't locked to Freebird's cloud — it routes through whatever backend you've configured, same as everything else. Set the backend to Ollama and get the full Pro feature set (multi-file edits, terminal commands, checkpoints, project memory) with every request staying on your machine. Built for teams whose policy forbids cloud AI, or freelancers working under an NDA that does the same — the model runs locally, so there's nothing to disclose.

Smart Chat with File Context

Type @filename to inject any file into the conversation. Type / to see all available commands.

Semantic Codebase Search (Pro)

Two ways the agent finds code: search_code for exact/keyword matches (like grep — good when you know the literal text), and semantic search for finding code by meaning — ask for "the payment retry logic" and it surfaces relevant code even if nothing is named "payment." Run Freebird: Build Codebase Index once per project to enable it; the agent picks whichever search fits your question automatically.

Git Integration

Generate commit messages, push to remote, and check git status from the chat panel — and the agent can call git status itself mid-turn when it's relevant to the change it's making, not just on request.

Project Rules

Write your own conventions to .freebird/rules.md — style preferences, things to always or never do, whatever you'd otherwise repeat in every prompt. Freebird loads it automatically into chat and Agent mode alike and follows it over its own defaults. Yours to write; Freebird only ever reads it. Use /rules to see what's loaded. Free and Pro both.

Project Memory (Pro)

Freebird saves notes about your project to .freebird/memory.md and loads them automatically. Use /memory to see what's saved and /forget to clear it. The difference from rules.md: this file is Freebird's own scratch notes, written opportunistically during Agent turns — rules.md is yours, written on purpose, and always takes precedence.

Share a Selection (Pro)

Select code, right-click → Freebird: Share Selection (or run it from the command palette), and get a link back — copied to your clipboard automatically. Whoever opens it sees just that selection in a plain read-only page, nothing else about your project: no repo access, no invite, no Freebird install required on their end. Links expire after 14 days. This is what "team collaboration" means here — not shared repo access, a scoped way to show a colleague one thing.

Related Locations — Lightweight Next-Edit Awareness (Pro)

After an Agent-mode edit, Freebird flags specific places elsewhere in your codebase that likely need the same change but weren't touched — another call site, a test asserting the old behavior, a doc describing it — as a short, dedicated card, not buried in prose. This is not Cursor's trained next-edit-prediction model; it's the agent using the same search tools it already has to double-check its own blast radius, and it only speaks up when it actually finds something.

Run Specialized Agents on Your Codebase

Map architecture, audit security, generate tests — all multi-file, all reversible. Or go freeform with BYOK + our managed backends.

Run Freebird: Use Prompt Template to start from a ready-made prompt — populates the chat input for you to edit before sending, rather than firing immediately. These are deliberately thorough (each is instructed to read broadly across your codebase — up to 15 tool calls in a turn) rather than fast, so expect a couple of minutes on a real project, not a quick chat-style reply. Three are free forever; 33 more (migrations, deeper reviews, compliance/accessibility, release automation, framework specialists, and more) are the Template Library — included with Pro/Enterprise/Team, or its own standalone subscription.

Codebase Cartographer — architecture overview, key abstractions and how data flows end to end, conventions worth knowing, real technical debt (not stylistic nitpicks), and a Mermaid.js dependency diagram. Pairs especially well with a large-context model like Kimi K3 on an unfamiliar codebase.

Security Auditor — cites the exact file and line for every finding, the concrete failure scenario (specific input/conditions that trigger it, not "this could be a vulnerability"), severity based on real exploitability, and a fix that doesn't change intended behavior. Fewer, real findings — no theoretical padding.

Multi-File Test Engineer — reads your existing test suite first to match its actual conventions, then prioritizes business logic with real consequences, easy-to-miss edge cases, and regression coverage for anything that's been a source of bugs before. Runs the tests it writes and fixes failures before finishing.

Pick the Right Model

BYOK models are free for everyone (bring your own API key/cost). Gemini Flash Lite and Ollama are also always free. Pro adds Agent mode (multi-file edits, terminal, checkpoints) and unlimited cloud edits on Claude Haiku 4.5 — automatically falls back to Gemini if Anthropic is ever unreachable, so Pro never hard-fails.

Using Freebird From Hong Kong or Mainland China

Google and Anthropic each independently restrict API access from Hong Kong and mainland China as their own policy — this isn't something Freebird can route around, and in Hong Kong's case it isn't related to local network censorship either (Hong Kong isn't behind mainland China's firewall; Google and Anthropic simply don't serve API traffic from the region). In practice: the free tier's default Gemini backend and Pro's default Claude backend may not work reliably, or at all, from either location, even with a valid subscription.

If you're in Hong Kong or mainland China, skip the cloud tier and use BYOK with DeepSeek or Qwen instead — both are Chinese-origin providers with no regional restriction, and DeepSeek in particular is priced competitively for what it offers. Ollama (fully local, nothing ever leaves your machine) also works everywhere, regardless of any provider's regional policy.

Getting Started

Option 1 — Just Install (Recommended)

1. Install Freebird AI
2. Open chat (Ctrl+Alt+O)
3. Start coding — 20 free AI edits/day, no setup needed

Option 2 — Ollama (Unlimited Free, Local)

1. Install Ollama
2. Run ollama pull qwen2.5-coder in a terminal
3. Run Freebird: Configure AI Backend → select Ollama

Option 3 — Anthropic Claude (BYOK, free)

1. Get an API key at console.anthropic.com
2. Run Freebird: Configure AI Backend → select Anthropic Claude

Option 4 — OpenAI (BYOK, free)

1. Get an API key at platform.openai.com
2. Run Freebird: Configure AI Backend → select OpenAI

Option 5 — DeepSeek (BYOK, free)

1. Get an API key at platform.deepseek.com
2. Run Freebird: Configure AI Backend → select DeepSeek

Option 6 — Qwen 2.5 (BYOK, free)

1. Get an API key at dashscope.console.aliyun.com
2. Run Freebird: Configure AI Backend → select Qwen 2.5

Option 7 — Kimi K3 (BYOK, free)

1. Get an API key at platform.moonshot.ai
2. Run Freebird: Configure AI Backend → select Kimi K3
3. Defaults to kimi-k3 (2.8T frontier, 1M token context) — set freebird.model to kimi-k2 for the previous generation instead

Option 8 — Custom Provider (BYOK, free)

Point Freebird at any OpenAI-compatible API — OpenRouter, Together, Groq, Fireworks, a self-hosted vLLM/LM Studio server, or anything else that speaks the /chat/completions format.

1. Run Freebird: Configure AI Backend → select Custom Provider
2. Enter the provider's base URL (e.g. https://openrouter.ai/api/v1) and the exact model id it expects (e.g. openai/gpt-4o)
3. Enter your API key when prompted

Commands

Chat Commands

How the Agent Works (Pro)

1. Reads your workspace file tree automatically
2. Fetches specific files it needs
3. Searches the codebase — exact/keyword matches via search_code, or by meaning via semantic search, whichever fits the question
4. Edits files with targeted diffs — Approve / Reject before anything changes
5. Creates new files — preview shown before creation
6. Runs terminal commands — shown before execution
7. Commits and pushes — requires your explicit approval
8. Checkpoints every turn that touched files — one click undoes the whole thing, anytime

Nothing is modified silently. You stay in full control.

Settings

Privacy

• Gemini Flash Lite (free tier): messages processed by Google's API under Freebird's own account. No code stored by Freebird.
• Claude Haiku 4.5 (Pro/Enterprise/trial cloud edits): messages processed by Anthropic's API under Freebird's own account, not yours — same trust model as the free tier's Gemini calls, just a different upstream provider. Automatically falls back to Gemini if Anthropic is unreachable.
• Ollama: all processing is local — no data leaves your machine. This applies to Agent mode too (Pro) — set the backend to Ollama and multi-file edits, terminal commands, and checkpoints all run without a single cloud call, which is what makes Freebird usable under a corporate no-cloud-AI policy or an NDA that forbids sending code off-machine.
• Anthropic / OpenAI / DeepSeek / Qwen / Kimi K3 / Custom Provider: code sent to that provider's API under your own account.
• Freebird AI (Ten Labs Pty. Limited) never collects or stores your code or conversation data — the one exception is Share Selection: a selection you explicitly choose to share is stored for 14 days (then auto-deleted) so the link keeps working, since the whole point is that the recipient doesn't need Freebird or repo access to view it.
• Web fetches (fetch_url): private/internal addresses (localhost, LAN ranges, cloud metadata endpoints) are blocked, and fetched page content is explicitly marked as untrusted before it reaches the model.
• Checkpoint restores can't write outside your workspace folder, even if a checkpoint record were somehow corrupted.

Support

support@ten-labs.com.au — payments, license activation, or anything else.

Contributing

Open source (MIT). Issues and PRs welcome at the GitHub repository.

License

MIT — Copyright © 2025 Ten Labs Pty. Limited