CodeWhale
MIT-licensed terminal coding agent that reads files, edits code, and runs commands with the model you choose.
If you already pay a model provider and want a coding agent that edits files, runs your tests, and stays on your machine, CodeWhale is the most configurable free option in the terminal-agent category — the runtime is MIT and $0, with DeepSeek as the documented default provider and Ollama, vLLM, and SGLang for fully local runs. The Fleet and Operate layers do genuinely useful multi-agent orchestration without a per-seat fee. The new /router Auto routing (Jev, provider fast tier, Off, or Custom) and default-on code mode add real orchestration. Pass if you need a polished GUI code review flow, a released desktop app, or a fully managed cloud runner — hosted web, desktop, and cloud computers
Verified 6d ago · liveness 75/100 · cite: rightaichoice.com/tools/codewhale
- Developers who want a coding agent that never sends the repository to a vendor cloud
- CLI-first engineers who want agent runs scriptable through codewhale exec in CI
- Shops that need per-role model routing so cheap models do exploration and stronger ones do edits
- Teams standardizing on a mix of local and hosted models under one workflow
- Non-technical users expecting a point-and-click GUI — this is terminal-first
- Teams that need a fully managed cloud runner with zero local setup — cloud computers are still in development
- Buyers who want a shipped desktop experience today — the macOS app is a development build with no public download yet
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip CodeWhale if you want a shipped desktop app or a managed cloud runner you can buy today — the hosted web app, macOS desktop build, and cloud computers are still development previews.
The runtime is $0, but every hosted-model task is billed directly by your provider — DeepSeek, OpenAI, Anthropic, or OpenRouter — so agent-heavy runs spend real money per token.
CodeWhale's runtime is free and MIT-licensed with a single $0 tier, so cost scales with your model provider rather than with seats. For a solo CLI developer running DeepSeek, that makes it cheaper than per-seat commercial terminal agents; teams running heavy hosted inference should compare their provider bill against those subscriptions before assuming the free runtime wins.
In short
CodeWhale — MIT-licensed terminal coding agent that reads files, edits code, and runs commands with the model you choose. Best for Developers who want a coding agent that never sends the repository to a vendor cloud, CLI-first engineers who want agent runs scriptable through codewhale exec in CI, Shops that need per-role model routing so cheap models do exploration and stronger ones do edits. Free to use.
What's new in CodeWhale
Checked 6 days agoAcross the latest 5 updates: 2 feature updates, 1 launch and 2 changelog entries.
CodeWhale v0.10.0 released
v0.10.0 was published Sep 22, 2026, and the source tree matches the published release.
Official model routing added with /router presets
/router (or /model router) configures an Auto router with Jev, provider fast tier, Off, or Custom presets; each preset runs one test call before saving, and /status shows choice, cost and latency.
Code mode composes MCP and plugin tools, on by default
execute_tools programs can call MCP tools, every nested call passes the same approval gate as a direct call, and each keeps its receipt. code_mode = false turns it off.
Delegated agent results no longer dropped when host is busy
A delegated agent's final result is never dropped when the host is busy, so a finished child no longer leaves a ghost Running row behind.
Stalled turns now report their phase instead of hanging
The turn loop records its phase and last progress, so an overdue phase surfaces instead of hanging silently until the stream idle timeout.
What people actually say about CodeWhale — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
11 mentions across 2 sources (Hacker News, GitHub) · researched Jul 3, 2026.
Average across the 2 sources that answered — each source counts once, not each post.
- +Cost-saving by routing low-complexity tasks to cheaper models like DeepSeek Flash.
- +Completely free and open-source (MIT license) with no subscription required.
- +Works with any AI model, prioritizing open-source and local runtimes.
- +Autonomous multi-step task planning and self-correction on failures.
- +Sandboxed execution via macOS seatbelt, Linux landlock, and Windows containment.
- −Steep learning curve for non-terminal-savvy users due to TUI/CLI interface.
- −High number of open GitHub issues (295) indicates ongoing instability.
- −Active security hardening work suggests potential unresolved vulnerabilities.
- −Some users switch between tools frequently, implying lack of stickiness.
- −Documentation and onboarding could be improved for beginners.
- • Cost of AI model API usage if using cloud models (e.g., DeepSeek API fees)
- • Compute resources for local models (GPU recommended)
Viability Score
How well maintained and how widely used is CodeWhale? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Terminal TUI with Plan, Work, and Operate modes
- Headless command execution via codewhale exec for scripts and CI
- Local browser client (codewhale web) connecting to the same on-computer session
- Runtime API that consumes and exposes MCP tools over stdio/HTTP
- Bring your own hosted model key via codewhale auth set --provider
- Local model support on vLLM, SGLang, and Ollama over localhost, usually keyless
- DeepSeek as the documented default model provider
- Official model routing via /router (or /model router) with Jev, provider fast tier, Off, and Custom presets
- Fleet: assign parts of a task to agents with different models and roles
- Operate mode runs structured workflows (Workflows, Lanes) with named phases and shared budgets
- Permission levels: Ask, Auto-Review, or Full Access
- Code mode (execute_tools) composing MCP and plugin tools, on by default and passing the same approval gate
- Computer Use plugin, explicit opt-in, notarized macOS build
- Saved sessions keeping conversation and tool results together for resume
- codewhale doctor for diagnosing provider connection and local tool issues
About CodeWhale
CodeWhale is an MIT-licensed coding agent that runs in your terminal. Point it at a project and it reads the repository, plans a multi-step task, edits files (showing each edit as a diff), runs commands and tests to check its own work, then reports what changed. The current published release is v0.10.0, published Sep 22, 2026. Model choice is per-session: connect a hosted provider such as DeepSeek (the default), OpenAI, Anthropic, or OpenRouter with your own API key, or run a local model on vLLM, SGLang, or Ollama on localhost, which usually needs no key. You pay the model provider directly; the CodeWhale runtime is $0. Work happens across four surfaces: a Terminal TUI with Plan, Work, and Operate modes; headless runs via codewhale exec for scripts and CI; a local browser client launched with codewhale web that connects to the same session on your computer; and a Runtime API that consumes and exposes MCP tools. Permissions are set separately from modes, with Ask, Auto-Review, or Full Access deciding how much the agent does before it stops to ask — in the default Ask posture it edits files inside the workspace immediately (showing the diff) but asks before running shell commands. For larger jobs, Fleet assigns parts of a task to agents with different models and roles, while Operate runs structured workflows (Workflows, Lanes) with named phases and shared budgets. An optional Computer Use plugin lets the agent see and operate other applications; it is an explicit opt-in that asks for system permissions. CodeWhale sits apart from managed cloud agents that assume a subscription and a vendor-hosted model: the source is on GitHub, the runtime is free, and the only bill is what your model provider charges you directly. The trade is a terminal-first workflow and a real setup step. The hosted web app, desktop app, and cloud computers are still development previews rather than released products.
Behind the Verdict
CodeWhale is a terminal-first, MIT-licensed coding agent. The core loop is honest and legible: it reads your repository, plans the task, edits files as diffs, runs commands (with a timeout on git_fetch, and git commands that no longer stop to ask for a password or host-key confirmation inside the terminal), and reports results. Modes and permissions are deliberately separate — Plan, Work, and Operate decide the kind of work, while Ask, Auto-Review, or Full Access decide when it stops to ask. Plan blocks file changes and shell execution, so you can explore a codebase safely, and the default Ask posture applies file edits inside the workspace immediately while asking before shell commands. Strengths. Bring-your-own-model is the real story. DeepSeek is the documented default, and the docs list OpenAI, Anthropic, OpenRouter, and local runners (Ollama, vLLM, SGLang) side by side. Local models usually need no API key, which matters if your repository cannot leave the machine. The runtime is free and the source is on GitHub, so the only bill is your provider's. Multi-agent work is real: Fleet saves roles and the model each role uses (codewhale fleet status counts queued, running, and finished runs), and Workflows run as Lanes with named phases and shared budgets. The v0.10.0 additions are substantive — /router sets up Auto routing with presets (Jev, provider fast tier, Off, Custom), each preset makes one test call before saving, and /status reports the router's choice, cost, and latency. Code mode is on by default: execute_tools programs can call MCP tools, each nested call passes the same approval gate as a direct call, and every nested call keeps its receipt. Rough edges that used to bite are being fixed — a busy host no longer drops a delegated agent's final result, a stalled turn now reports its phase instead of hanging, and upgrading no longer disables the built-in Computer Use bundle when capabilities are unchanged. Weaknesses. This is not sign-up-and-go, and the docs say so: you must install a binary, put ~/.local/bin on your PATH, and connect a model before anything useful happens. v0.10.0 does not warn you if you have no key configured on your first message (the docs tell you to press F3). The 60-second quickstart covers Linux and macOS; Windows, macOS, and Android are mostly out of scope of the verified install guide, and Android/Termux is a preview. Telemetry is on by default (PostHog aggregate usage counts) until you run codewhale config set telemetry false or set CODEWHALE_TELEMETRY=0. The local browser client is loopback-only. Advanced features — sub-agents, per-sub-agent provider routing, OS sandboxing (bubblewrap on Linux), and Fleet/WhaleFlow durable workflows — require configuration and understanding of the architecture. The hosted web app, desktop app, and cloud computers are development previews, not products you can buy today. Where it fits. CLI-first engineers and small teams who already hold provider keys and want agent runs
Researching CodeWhale? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas CodeWhale actually fits — and what changes day-one when you adopt it.
Install with curl -fsSL https://codewhale.net/install.sh | sh, put ~/.local/bin on the PATH, save a DeepSeek key with codewhale auth set --provider deepseek, then run codewhale in the project folder and start in /mode plan to have the agent explain the repository.
Outcome: You understand an unfamiliar codebase without any file changes, then switch to /mode work and have the agent edit files as diffs and run the tests before you approve each shell command.
Run codewhale exec from a CI script for a one-shot task, or drive threads and approvals over the local HTTP Runtime API, with local Ollama as the model so the repository stays on the runner.
Outcome: Agent runs become a repeatable CI step with no vendor cloud in the loop and no per-seat fee.
Use /fleet setup to save roles and the model each role uses, giving a cheap model the exploration work and a stronger one the edits, then check codewhale fleet status from the shell to see queued, running, and finished runs.
Outcome: A large task is split across agents under one workflow, with progress visible from any terminal and provider costs routed to the cheaper model where it matters.
Use Cases
- Automate refactoring and bug fixing across large codebases via multi-step planning
- Run code reviews and enforce standards with constitution-based agent guidance
- Execute repeatable Workflows as Lanes, or batch tasks, with Fleet orchestration
- Develop and test API integrations with MCP tools exposed over stdio or HTTP
- Build custom skill libraries for domain-specific coding assistance
- Run one-shot jobs in CI or drive threads and approvals over the local HTTP API
- Use local models on Ollama, vLLM, or SGLang for offline code analysis
- Propose and confirm a cloud agent that opens a pull request (preview)
Models Under the Hood
as of 2026-09-22
Limitations
- CodeWhale is a terminal-first open-source coding agent (v0.10.0, MIT) for macOS, Linux, and Windows that needs a model connection: connect a hosted provider key or run a local model, which usually needs no key.
- The verified install guide covers Linux and macOS; macOS, Windows, and Android are largely out of scope there apart from notes, and Android/Termux is a preview.
- The local browser client is loopback-only.
- Telemetry sends aggregate usage counts (PostHog) by default until you turn it off with codewhale config set telemetry false or CODEWHALE_TELEMETRY=0.
- The hosted web app, desktop app, and cloud computers are development previews, not released products.
- Advanced features — sub-agents, per-sub-agent provider routing, sandboxing, and Fleet/Workflow/Lane durable orchestration — require configuration and understanding of the architecture.
- Code mode is documented as experimental.
as of 2026-10-03
Verification history
We have re-verified CodeWhale 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published CodeWhale tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Open Source (MIT)
$0
Ideal for
Solo developers and CLI-first teams who already hold a model provider key or can run a local model, and want an agent runtime with no per-seat fee.
What this tier adds
Starting tier: the full CodeWhale runtime under MIT at $0, with model usage billed directly by your chosen provider.
Where the pricing makes sense
The company stage and team size where CodeWhale's pricing actually pencils out — and where peers do it cheaper.
CodeWhale's runtime is free and MIT-licensed with a single $0 tier, so cost scales with your model provider rather than with seats. For a solo CLI developer running DeepSeek, that makes it cheaper than per-seat commercial terminal agents; teams running heavy hosted inference should compare their provider bill against those subscriptions before assuming the free runtime wins.
Setup time & first value
How long it actually takes to get something useful out of CodeWhale — broken out by persona, not the marketing-page minute.
Solo developer on Linux or macOS: about 60 seconds to install and roughly 5 minutes to save a provider key and run a first task. Air-gapped or security-reviewed machines: longer, following the manual download path from GitHub Releases. Local-model users: add the time to stand up Ollama, vLLM, or SGLang, which needs no key but does need the machine.
Switching to or from CodeWhale
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From GitHub Copilot Workspace: install the CodeWhale binary, connect DeepSeek or another provider key with codewhale auth set --provider, then run codewhale in your repository and start in /mode plan.
- →From a subscription terminal agent: keep your existing provider key where possible, save it in CodeWhale, and move repeatable jobs to codewhale exec or the Runtime API.
- →From local Ollama scripts: point CodeWhale at the same localhost Ollama endpoint and use the TUI for planning and diffs instead of hand-rolled prompts.
- →From manual CLI workflows: add your project's conventions as a repository constitution so the agent follows your standards.
- ↗To a managed cloud agent: export your task list and re-create the workflows in the vendor's hosted runner, since CodeWhale's Fleet and Workflow definitions are local to your config.
- ↗To a GUI-first code review tool: review diffs in CodeWhale first, then push and use the other tool for graphical review, since CodeWhale review is terminal-based.
- ↗To another BYOK terminal agent: keep your provider key, since CodeWhale's connection is a plain provider key saved in the local secret store.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “CodeWhale”, and we withheld 6: 6 could not be judged, because “CodeWhale” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about CodeWhale.
Official links
Tools that pair well with CodeWhale
Common stack mates teams adopt alongside CodeWhale, with the specific reason each pairing earns its keep.
Cline
Open-source coding agent that reads your repo, edits files, and runs terminal commands across VS Code, JetBrains, CLI, and a desktop app.
Claude Code
Claude Code is Anthropic's agentic coding assistant that plans, edits, and runs commands across your repo from the terminal, IDE, or browser.
Openclaude
Open-source terminal coding agent that runs against any model you point it at — OpenAI, Gemini, Codex, Grok, Ollama, LM Studio and 200+ more.
Featured Head-to-Head Comparisons
Codewhale vs Locus Robotics
These tools serve completely different domains: Locus Robotics is an industrial warehouse automation platform for physical fulfillment, while CodeWhale is a local terminal coding agent for software developers. Choose Locus if you need to boost warehouse picking productivity 2-3x with AMRs and RaaS; choose CodeWhale if you want a free, open-source, model-agnostic AI coding assistant that runs on your command line.
Codewhale vs Truleo
Buyers should not confuse these two tools — they serve entirely different domains. If you are a law enforcement agency looking to unify siloed data and automate investigative leads, Truleo is purpose-built for you. If you are a developer wanting a free, local, multi-model terminal coding agent, CodeWhale is the clear choice. They are not competitors but complementary in the broadest sense.
Codewhale vs Presto Voice
CodeWhale and Presto Voice serve completely different domains. CodeWhale is a free, open-source coding agent for developers who want local, model-agnostic automation. Presto Voice is an enterprise voice AI for QSR drive-thrus, with recent high-profile partnerships like Dairy Queen. Choose based on your industry: developers should pick CodeWhale; QSR operators should explore Presto Voice.
Alternatives to CodeWhale
View allCline
Open-source coding agent that reads your repo, edits files, and runs terminal commands across VS Code, JetBrains, CLI, and a desktop app.
Claude Code
Claude Code is Anthropic's agentic coding assistant that plans, edits, and runs commands across your repo from the terminal, IDE, or browser.
Openclaude
Open-source terminal coding agent that runs against any model you point it at — OpenAI, Gemini, Codex, Grok, Ollama, LM Studio and 200+ more.
Frequently Asked Questions
Categories
Used CodeWhale? Help shape our editorial sentiment research.