AI Tools.
Choose with evidence.

Why consider AdaL

• Coding tasks that delegate to specialized workers.
• Combining skills, hooks and local or hosted model connections.

What to weigh up

• Local-model support is preview and excludes some capabilities such as image input.
• Included credits, provider subscriptions and BYOK have different billing paths.

Models & access

Multiple hosted providers, third-party subscription access and local Ollama models are documented; local support is preview.

Included credits and paid/BYOK options Plan credits and your selected model-access option determine costs; check the current Pricing & Usage terms.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider AIChat

• Combine shell assistance, document retrieval and tool use in one CLI.
• Use hosted and local models without switching the interface.

What to weigh up

• RAG and tool integrations require configuration.
• Role/agent templates are not evidence of a complete autonomous coding workflow.

Models & access

Supports hosted providers, Ollama and compatible API endpoints through one configurable client.

BYOK or locally served models Configure a hosted model provider using your own account, or connect to Ollama or another compatible local service. Hosted usage follows the provider's rates and limits; local operation uses your own hardware. The client does not establish one bundled inference price across providers.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Aider

• Make scoped code changes with repository maps and Git integration.
• Choose a preferred hosted or local model instead of a vendor-managed catalog.

What to weigh up

• Model compatibility and edit reliability vary; read the model-specific warnings.
• Skills/MCP/subagent parity with larger agent platforms was not established.

Models & access

Supports many hosted providers and local endpoints such as Ollama and LM Studio. Model-specific warnings and settings are documented.

Provider API usage or local compute Connect your own cloud-provider credentials or run a compatible local model. Some providers offer limited free API access; others meter model usage. Aider's client license does not include a hosted model allowance, and self-hosting still requires your own compute resources.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider AiderDesk

• Review context, token use and proposed changes in a dedicated app.
• Delegate tasks to specialized subagents while keeping an existing editor.

What to weigh up

• Model providers and approval policies still need configuration.
• AiderDesk is a separate application, not the Aider terminal interface.

Models & access

Cloud providers and Ollama are documented. This is an orchestration app distinct from Aider CLI.

Open-source client; model costs separate Provider usage or local hardware supplies inference.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Amp

• Move a task between local CLI, cloud Orbs and native apps.
• Use shared threads and workspace model routing.

What to weigh up

• Model routing and gateway access have plan restrictions.
• Cloud machine usage and agent usage have separate allowances.

Models & access

Uses multiple model families. BYOK and linked subscriptions are supported; compatible gateway connections remain early access on selected paid tiers. Local-model operation was not verified.

BYOK, subscriptions or usage billing Model/tool usage and cloud Orbs are billed separately; workspace and tier restrictions apply.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Antigravity

• Work with code, browser actions and review artifacts in one environment.
• Use documented skills, subagents and hooks to customize agent work.

What to weigh up

• Model availability changes by plan and enterprise policy.
• Local-weight execution was not established for the IDE in this review.

Models & access

The IDE offers a plan-dependent hosted model selector including Google, Anthropic and open-weight models. An open-weight model in the selector does not establish on-device inference.

Plan quotas and AI credits Free and paid Google AI plans have different model availability and quotas.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Antigravity CLI

• Use Antigravity from SSH, tmux or a terminal workflow.
• Carry settings and conversations into the Antigravity visual environment.

What to weigh up

• Model availability changes by plan and enterprise policy.
• The CLI is a separate surface from the Antigravity IDE; one is not the installer for the other.

Models & access

Antigravity offers a plan-dependent catalog spanning Gemini and selected third-party models. CLI also documents direct Gemini API-key authentication; arbitrary local-model support was not established.

Included quota; paid plans and overages Google-account access includes a baseline quota. Google AI paid plans increase allowances, and eligible subscribers can purchase AI credits for optional usage beyond that quota. Model access and limits depend on the plan. Direct Gemini API-key authentication is a separate CLI access path.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider AnythingLLM

• Build document-grounded workspaces with citations and agent tools.
• Choose a desktop installation or a self-hosted multi-user instance.

What to weigh up

• Multi-user permissions and embeddable widgets are documented for Docker, not desktop parity.
• Local privacy depends on the selected model, embedding and other providers.

Models & access

Connects local runtimes including Ollama and LM Studio or cloud providers. Supports document ingestion, agent workflows and configurable model routing.

Open-source self-hosting; hosted option Hosting and chosen provider usage are separate costs.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider AstrBot

• Chat-platform assistants using your own model service.
• Web chat with plugins and isolated code execution.

What to weigh up

• Requires deployment and configuration of model and messaging connections.
• Desktop deployment is explicitly discouraged for server use.

Models & access

Official provider list includes hosted APIs, OpenAI-compatible services, Ollama and LM Studio.

Self-hosted software Model-service and hosting costs depend on your deployment.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Augment

• Use an indexed codebase context service inside VS Code or JetBrains.
• Run development automation from triggers and schedules through Cosmos.

What to weigh up

• Token, service-fee and cloud-compute charges can all contribute to cost.
• Subagents and hooks are documented for CLI/Cosmos; do not assume full extension parity.

Models & access

Hosted model selection spans multiple providers; the last selection or organization default is used. Local model support was not confirmed.

Tokens, service fee and cloud compute Current Cosmos/Auggie pricing includes model usage, a service fee and applicable compute; see product-specific terms.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Autohand Code

• Use local inference options or hosted providers from a customizable terminal agent.
• Extend workflows through skills, hooks, MCP and agent definitions.

What to weigh up

• LICENSE is Apache-2.0, but README and COMMERCIAL.md impose a revenue-based commercial restriction.
• The licensing conflict needs clarification before calling the runtime unrestricted open source.

Models & access

Documents hosted providers, Ollama, llama.cpp and MLX, with provider switching.

Provider costs; commercial terms vary Hosted model providers bill through your configured accounts; local backends use your own compute. The repository describes no-charge use for some users and a commercial license requirement for certain organizations, but its licensing notices conflict. Check both notices before assuming organization-wide client use is free.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Bolt

• Turn a design or prompt into an editable web application in the browser.
• Use built-in hosting/database workflows or import an existing GitHub project.

What to weigh up

• Focuses on supported web stacks; it is not a general local shell agent.
• Token use, hosting and production services need separate cost review.

Models & access

Hosted agent selection is documented; no arbitrary local inference endpoint verified.

Token-based plans and hosted services Build usage is token-based; deployed application services can have separate plan constraints.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider ChatGPT

• Research sources, analyze uploaded material and produce documents or spreadsheets in a managed workspace.
• Use desktop file/app access and inspect finished artifacts; use web Work for cloud tasks that can continue after closing the app.

What to weigh up

• Features vary by plan, rollout, region and workspace policy; Work and Codex share usage.
• Local task execution and local MCP access do not establish local model inference or an offline assistant.

Models & access

Chat and Work expose plan-dependent OpenAI models. Desktop Codex views also share Codex configuration, but their provider options are not attributed to ordinary Chat or voice.

Free access and paid plans; usage limits apply Work and Codex share usage and credits. API access is a separate billing path.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider ChatWise

• Keep a desktop chat workspace while switching among cloud APIs and separately served local models.
• Use Agent preview with project files, portable SKILL.md packages and delegated exploration.

What to weigh up

• Agent preview explicitly disables MCP and Artifacts and currently has no sandbox.
• Normal chat MCP lacks OAuth and server-initiated Streamable HTTP messages; docs also warn local Ollama tool use is not optimized.

Models & access

Bring provider keys or connect compatible endpoints, including Ollama and LM Studio. Requests go to the chosen model provider.

Personal free tier; paid Pro license plus model usage Free access is limited to personal non-commercial use. Pro is a one-time license with a bounded feature-update period; model access is separate.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Cherry Studio

• Compare several model responses inside one desktop conversation workflow.
• Use custom assistants and documents across local and cloud providers.

What to weigh up

• Community and commercial editions have different terms.
• Cloud calls and integrations have different privacy boundaries from local inference.

Models & access

Cloud providers plus local Ollama and LM Studio connections; multi-model simultaneous conversations are documented.

Community client; commercial edition The community edition is AGPL; provider use and commercial-license terms are separate.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Claude

• Connect work services and create editable reports, spreadsheets and presentations.
• Use Desktop for access to permitted local folders, local MCP tools and supported computer workflows.

What to weigh up

• Unified chat/Cowork is rolling out by plan and account; older accounts still show separate modes.
• Remote MCP originates from Anthropic cloud and needs a publicly reachable server; local desktop extensions have a different connection path.

Models & access

Official consumer plans offer Claude models. Optional externally configured desktop routing is not verified as a general feature here; local files and connectors do not imply local model weights.

Free access and paid subscriptions Usage, models and advanced work features depend on plan; enterprise terms differ.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Claude Code

• Use Claude with project instructions, skills, MCP and specialized agents.
• Apply lifecycle hooks and organization-managed settings to coding workflows.

What to weigh up

• Official model choice is Claude; cloud hosting options are not unrestricted model choice.
• Included usage and defaults depend on the account and provider.

Models & access

Official model configuration targets Claude through Anthropic or supported cloud deployments. Changing ANTHROPIC_BASE_URL changes the endpoint, not the model family; runtime defaults depend on account/provider.

Claude plan or metered API The sign-in method determines billing. Claude subscriptions and organization seats include usage allowances, with optional usage credits on eligible plans. Anthropic Console and supported cloud-provider access are billed by token to the corresponding account; a Claude subscription does not fund separate API-key usage.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Cline

• Keep an existing editor while choosing model providers and local inference.
• Use approval controls, MCP and reusable skills across documented Cline surfaces.

What to weigh up

• Inference cost depends on provider, model and work performed.
• Desktop availability differs from CLI: official docs list macOS and Windows beta; Linux Desktop is not confirmed.

Models & access

BYOK cloud providers, OpenAI-compatible endpoints and local runtimes including Ollama and LM Studio. Cline-hosted usage and ClinePass are alternative billing paths.

BYOK, usage billing or ClinePass The open-source client does not make paid inference free. Promotional free models can change.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider CodeArts Agent

• Use requirements, implementation design and task planning as explicit coding stages.
• Assess centralized skills and model management for a Huawei Cloud organization.

What to weigh up

• Features, token allowances and governance depend on plan.
• Regional download and plan pages differ; verify the intended service region.

Models & access

A managed model selector is documented. Enterprise plans list custom model management, but on-device inference is not verified.

Plan seats and token allowances Free and paid editions have different allowances and enterprise capabilities.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider CodeBuddy

• Move from design/plan to implementation with IDE integrations.
• Configure external services through MCP and custom API models.

What to weigh up

• Promotional credits and feature allowances can expire.
• CodeBuddy, its CLI and WorkBuddy have distinct surfaces; do not mix their pricing tables.

Models & access

IDE supports custom model API integration; the fetched overview does not establish on-device weight execution.

Free allowance and subscription credits Pricing lists plan-specific credit and model allowances, including time-limited promotional benefits.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Codemaker

• No source-verified recommendation is available for this record yet.

What to weigh up

• Do not choose on the basis of this incomplete registry entry.
• A similarly named CodeMaker AI service documents IDE plugins, but its identity has not been tied to this registry key.

Models & access

Model access has not been established from the sources reviewed.

Not independently verified Check the official service terms before choosing a plan.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Codex

• Work across local terminal, editor and cloud coding sessions.
• Customize the open-source CLI, provider configuration, skills, subagents and trusted hooks.

What to weigh up

• The CLI license does not grant free access to hosted models.
• Custom endpoints must match the documented API capabilities.

Models & access

OpenAI is the standard account path. Advanced CLI configuration supports custom model providers and local Ollama-compatible endpoints; compatibility is provider-specific, not proof every feature works with every model.

ChatGPT allowance or API usage ChatGPT sign-in uses plan allowances shared with ChatGPT Work; eligible plans can extend usage with credits. API-key mode bills the API account and excludes hosted features such as cloud code review. Configured local or third-party providers have their own inference or compute costs.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Command Code

• Use hosted open models or direct compatible provider/local endpoints.
• Apply reusable taste preferences, skills and hooks across coding tasks.

What to weigh up

• BYOK is explicitly beta.
• Vendor claims of best harness performance are not independent comparative evidence.

Models & access

BYOK beta accepts compatible OpenAI or Anthropic APIs, including local Ollama/vLLM servers. Subscription connections are also documented; routes and usage depend on the provider.

Plan credits, BYOK or linked subscriptions Hosted access uses plan credits with rolling limits and optional on-demand top-ups. BYOK, compatible local endpoints and supported linked subscriptions provide separate routes. The documented local-only mode skips Command Code billing reads and routes usage through your chosen provider; provider terms and compute costs still apply.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Continue

• Inspect or reuse an Apache-licensed assistant codebase.
• Maintain an existing installation with a clear awareness of upstream status.

What to weigh up

• Official repository is read-only and no longer actively maintained.
• Do not promote as a default new-user recommendation without a maintained successor.

Models & access

Documentation supports configurable providers and local Ollama/self-hosted models. Upstream is no longer maintained.

Open-source client; inference separate Current commercial service availability was not established; do not reuse historical subscription claims.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Corust Agent

• No source-verified recommendation is available for this record yet.

What to weigh up

• Official website returned502 during this pass.
• Public repository provides release metadata/license but no product README or agent implementation source.

Models & access

No primary model-selection documentation retrieved.

Not established Current account and billing terms are unavailable from the reachable product documentation.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Crow

• Resuming coding sessions with searchable local history.
• Using one agent through terminal and ACP-native editor tools.

What to weigh up

• Web search requires the documented SearXNG/Docker backing service.
• MCP tool names are not namespaced, so added servers may collide.
• WSL2 is recommended for Windows setup.

Models & access

Requires an OpenAI-compatible provider API key; custom endpoints are documented, but local-weight hosting was not explicitly established.

Open-source client; provider access separate Provider credentials and backing-service hosting are required as configured.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Crush

• Use a terminal-focused interface with LSP context.
• Switch compatible models while preserving session context.

What to weigh up

• Source is available under FSL with a competing-use restriction; it is not an unrestricted open-source license.
• Hooks are explicitly described as preliminary.

Models & access

Supports multiple providers and custom OpenAI- or Anthropic-compatible APIs. Local inference depends on endpoint compatibility and was not independently exercised.

Hyper plan, BYOK or local compute Optional Charm Hyper offers a free tier and paid subscriptions. Alternatively, connect your own provider API keys or compatible local models. External inference and local compute costs depend on that choice; the client's source-available license is a separate consideration.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Cursor

• Keep code review, planning, editing and agent sessions together.
• Customize the agent with documented rules, skills, hooks and MCP.

What to weigh up

• Model and usage allowances depend on the chosen plan.
• Cloud execution and editor execution have different data and environment boundaries.

Models & access

The managed catalogue spans multiple providers. Local-weight execution was not verified. A model selector is different from arbitrary local endpoints.

Plans and metered model usage Use the current model/plan table; no fixed cost per task is implied.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Deep Agents Code

• Terminal coding with memory, subagents and model choice.
• Using configurable tool backends or remote sandboxes.

What to weigh up

• The coding product and Python/JavaScript SDKs are distinct surfaces.
• Windows-specific installation was not established in this pass.

Models & access

Official SDK README explicitly supports tool-calling frontier APIs and self-hosted models via Ollama, vLLM or llama.cpp; Code docs say models/providers can be switched.

Open-source client; model access separate Provider and optional hosted observability/sandbox costs are not included by the client license.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Devin CLI

• Start on local code and hand longer work to Devin Cloud.
• Use managed model selection with team configuration.

What to weigh up

• CLI and cloud Devin are separate tools.
• CLI currently does not inherit Devin Knowledge, Playbooks or Secrets.

Models & access

Offers a managed catalog from several model vendors plus Adaptive routing and Fusion pairings. Enterprise teams can restrict the catalog; arbitrary BYOK/local endpoints were not verified.

Plan quota plus optional usage credits Paid self-serve plans include CLI usage within allowances shared across Devin sessions and desktop. On-demand credits extend usage; team plans pool those credits across members. Enterprise contracts instead use Agent Compute Unit billing. Review the selected seat and plan because entitlements differ.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Devin Desktop

• Move existing Windsurf projects into the documented successor IDE.
• Use built-in review, browser previews and local agent work.

What to weigh up

• A Devin account is required.
• Legacy Cascade and Devin Local have different model availability; extension compatibility also differs from VS Code.

Models & access

Hosted model choices include different model families. Some models require Devin Local and are unavailable in legacy Cascade. Local agent execution is not proof of local model inference.

Free tier; paid subscriptions and extra model usage Devin Free limits agent quota and model access. Paid plans expand both and allow extra usage purchases. Team plans combine an organization fee with developer-seat charges; enterprise controls are separately contracted.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Dexto

• Coding through either a CLI or web UI.
• Building configurable agents with local models and MCP tools.

What to weigh up

• Source is available under Elastic License 2.0, which restricts offering the software as a managed service.
• Official README labels the project beta.

Models & access

Hosted providers, gateways, Ollama and local GGUF models are documented.

Source available; model access separate Model-provider charges and Elastic License terms apply; review the terms for your deployment.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider DimAgent

• Using local or hosted models across a terminal and desktop coding workspace.
• Forking tasks to background agents with configurable models.

What to weigh up

• Public GitHub repository is an issue tracker, not product source.
• Web search requires a separate Serper key; planned features should not be assumed shipped.

Models & access

Official docs explicitly support custom endpoints, Ollama and LM Studio in addition to hosted providers.

BYOK or local models; Coding Plan coming soon Hosted model costs depend on your connected provider; local inference uses your own hardware. The Coding Plan is listed as coming soon.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Dirac

• Long coding tasks with separately configured subagents and user steering.
• Code inspection and edits using syntax-tree and stable-anchor tools.

What to weigh up

• Structural editing depends on parser/index coverage and requires verification for dynamic references.
• Custom tool creation and automatic approvals depend on explicitly selected permissions/configuration.

Models & access

README describes provider choice and bring-your-own credentials with shared configuration across interfaces; local weights were not explicitly verified.

Open-source client; provider access separate Costs depend on the model provider you connect.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Droid

• Use company-specific models or local endpoints from CLI and desktop.
• Package team skills, hooks and specialist droids.

What to weigh up

• Model flexibility is surface-specific.
• Factory warns that arbitrary custom models are not guaranteed to work out of the box.

Models & access

CLI and desktop support BYOK, generic compatible endpoints and local Ollama/vLLM-style servers. Custom models do not appear on Factory-hosted web/mobile surfaces.

Subscription allowance; extra usage and BYOK Individual subscriptions include the app, CLI and SDK with rolling usage limits. Optional prepaid Extra Usage extends access. BYOK has an included allowance, then plan-specific Factory charges can apply in addition to the model provider's charges; it is not an unlimited free platform route.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Eve

• Building custom agents from files and typed tools.
• Adding schedules and channel integrations to an agent application.

What to weigh up

• Framework and project scaffold rather than an end-user coding assistant.
• Official README states beta terms and possible API/behavior changes.

Models & access

AI Gateway model configuration is documented; arbitrary endpoint and local model support were not established.

Framework; service costs separate Configured model services and deployment infrastructure have separate costs.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider fast-agent

• Interactive coding with shell tools, skills and MCP servers.
• Composing agent workflows and using the same setup through ACP.

What to weigh up

• Shell support and built-in subagents are enabled by CLI options; capabilities depend on configuration.
• Windows filesystem/Docker MCP setups need documented configuration adjustments.

Models & access

Supports hosted providers and generic local endpoints; README explicitly demonstrates local llama.cpp configuration.

Open-source client; provider access separate Hosted model services and local inference hardware have separate costs.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Firebender

• Use Android/JetBrains context, refactoring tools and Compose previews.
• Connect a team model gateway through documented LiteLLM configuration.

What to weigh up

• BYOK and on-demand usage require an active paid plan.
• Some tiers add a tool fee alongside model inference costs.

Models & access

Hosted model selection, BYOK, Amazon Bedrock and organization-managed LiteLLM are documented. Generic on-device model support is not established by the gateway integration alone.

Subscription plus model usage Included usage, extra usage, BYOK availability and fees differ by tier; bonus capacity is not guaranteed.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider ForgeCode

• Send coding tasks directly from a configured Zsh prompt.
• Choose cloud or local providers and tailor agent definitions.

What to weigh up

• Zsh integration and terminal font setup add onboarding steps.
• Windows uses WSL or Git Bash in the documented install path.

Models & access

Official docs support cloud models, open-weight models and local models with provider configuration.

Provider-funded; optional Forge services Connect a provider API key, supported provider account or local endpoint. Model usage and local compute follow that choice. Optional ForgeCode Services add hosted context and guardrail services; signing into those services is distinct from configuring the model provider.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider fx

• Shell-oriented terminal coding and one-shot script tasks.
• Embedding the same agent core through libfx or ACP.

What to weigh up

• Official README labels fx experimental.
• Official install documentation lists macOS/Linux only; Windows support not established.

Models & access

Supports Vercel AI Gateway, supported subscriptions, and custom OpenAI-compatible endpoints including Ollama.

Open-source client; model access separate Model gateway, provider or subscription access is separate from client license.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Gemini CLI

• Use Gemini from a terminal with shell, search and MCP tools.
• Extend an open-source CLI with skills, hooks and editor integration.

What to weigh up

• Some advanced capabilities are marked experimental in the docs.
• Google authentication tier and API route determine available usage.

Models & access

Documented primary model and authentication paths use Gemini. This pass did not verify arbitrary local inference support.

Free allowance, subscription or metered API Billing depends on authentication. Personal Google sign-in has an included allowance; eligible Google AI and Gemini Code Assist subscriptions offer higher limits. Gemini API keys and Vertex AI have their own free or usage-based tiers. CLI requests and tool work consume the selected route's quota.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider GitHub Copilot CLI

• Work with GitHub issues and pull requests from the terminal.
• Use Copilot plans and organization policy across local and cloud agent tasks.

What to weigh up

• Organization users need the Copilot CLI policy enabled.
• Local/cloud sandbox features are documented as preview.

Models & access

Managed model selection spans several vendors. This is not evidence of arbitrary local endpoints; organization policies can constrain access.

Copilot plan usage Available with Copilot plans; organization policy and plan limits apply.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider goose

• Use the same general-purpose agent from desktop or terminal.
• Connect local models and MCP tools for research, files and development.

What to weigh up

• Model setup and extension configuration remain your responsibility.
• A desktop install and CLI install are separate choices.

Models & access

Supports hosted API providers and Ollama. ACP integrations can use supported existing coding subscriptions; inference terms remain with the provider.

Provider APIs, linked plans or local compute Configure a hosted provider API key, a supported subscription-backed integration or a local model runner. Goose does not supply a universal model allowance: billing and limits follow the selected provider or linked account. Local models use your own machine or server resources.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider GPT4All

• Start local document chat without a provider account.
• Use a desktop client or integrate local inference through Python.

What to weigh up

• Available model capabilities and hardware still constrain results.
• A local chat client is not a drop-in autonomous coding agent.

Models & access

Runs downloadable local models on desktops; the project also exposes a Python client. No API key is required for local model execution.

Open-source client; local inference Model licenses and hardware costs remain separate.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Grok Build

• Use the terminal TUI or headless CLI with a configured model ID and API base URL.
• Delegate repository exploration and planning to separate child contexts with built-in types that cannot run shell commands or edit files.
• Extend the terminal workflow with skills, plugin marketplaces, MCP servers and lifecycle hooks.
• Inspect and customize the published Apache-2.0 CLI/runtime source.

What to weigh up

• Sandboxing is off by default. Documented isolation uses Landlock on Linux and Seatbelt on macOS; enable a profile before relying on filesystem limits.
• Child-process network restrictions apply on Linux only; they do not block model API calls or built-in web tools.
• Built-in sandbox profiles do not permanently protect credential paths such as ~/.ssh; the docs advise a custom deny list. Session storage remains writable.
• Skill allowed-tools metadata does not enforce tool access, and general-purpose subagents retain full capabilities.
• The public repository is periodically synchronized from the vendor’s internal repository.
• The open-source client licence does not include hosted inference; account and provider billing still apply.

Models & access

The getting-started guide demonstrates grok-4.7 on the xAI API and documents custom model IDs and base URLs. It does not establish a universal Grok Build default. Local model serving was not separately verified.

Metered API or configured provider The xAI API-key route uses prepaid API credits or approved invoiced billing. Custom model endpoints use their own provider accounts and terms. Browser sign-in is also documented, but this review did not establish a bundled subscription allowance for that route.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Harn

• Building custom agents whose tool permissions and execution history are part of application code.
• Testing workflows with recorded model responses and deterministic replay.

What to weigh up

• A programming language/runtime, not a ready-made coding assistant.
• Pre-release language and CLI can change between releases.

Models & access

Primary README and getting-started guide explicitly support hosted and local models, including Ollama/llama.cpp/MLX/vLLM routes.

Open-source runtime; model costs separate Offline recorded-response demos require no provider; live hosted calls require credentials and may incur charges.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Hermes Agent

• Run recurring research and automation with persistent memories and skills.
• Continue work through supported messaging channels and remote execution backends.

What to weigh up

• Persistent memory and remote tools need intentional configuration.
• External models, hosting and optional tool services have their own costs.

Models & access

Supports Nous Portal, external APIs, local Ollama/vLLM and compatible endpoints. The local server must meet the tool-use context requirements documented by Hermes.

Optional Nous plan, BYOK or local compute Run the agent with external API keys or a locally served model, paying the chosen provider and any hosting costs. An optional Nous Portal subscription bundles model access and supported tool services. Separately connected search, browser or other tool accounts retain their own terms.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider IBM Bob

• Review an implementation plan before allowing edits.
• Use explicit subagent approvals and configurable external tools.

What to weigh up

• Old IDE releases are scheduled to stop working; use the current official installer.
• MCP servers are not preinstalled and require configuration.

Models & access

Bob uses managed model orchestration. The fetched FAQ rendered questions without answers, so model choice and local inference remain unverified.

Subscription with Bobcoin allowance; limited free trial Individual tiers include Bobcoins and Bob Shell access. Enterprise adds centralized user and coin management. Specialist modernization packages are separate paid add-ons, and their trials require a subscription.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider iFlow CLI (retired)

• Identify a legacy installation and locate migration guidance.

What to weigh up

• Official README announces shutdown on April 17, 2026 (UTC+8).
• Do not recommend new installations or advertise former free-model access.

Models & access

Historical provider access is not evidence of current availability after the announced shutdown.

Retired; former free access unavailable The official repository announces the service shutdown. Historical free-access claims and old usage allowances should not be treated as an available offer. Review the shutdown and migration information before relying on an existing installation.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider inference.sh / Belt

• Calling hosted AI tools and deploying durable agents.
• Distributing skills and connecting existing coding-agent harnesses.

What to weigh up

• Cloud platform account and usage limits apply.
• Belt orchestration and tool access should not be presented as a standalone local coding assistant.

Models & access

Cloud model catalogue and BYOK are described; this does not establish arbitrary endpoints or local weights.

Cloud usage/account billing Your platform plan, usage and connected model providers determine costs; check current account terms.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Jan Agent

• Run folder-based file and shell work using an existing Jan model setup.
• Reuse skills, MCP and provider settings with Jan Desktop.

What to weigh up

• Official docs label this a preview and the dev installer uses nightly builds.
• Local model tool reliability is still model-dependent.

Models & access

Uses cloud providers or OpenAI-compatible endpoints, including a local model served by Jan Desktop.

Bring your own inference; preview client fee unverified Uses configured cloud-provider credentials or your own compatible model server, including Jan Desktop. No separate Agent client fee was stated in the reviewed official overview/provider docs; provider-account charges and local hardware requirements remain separate.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Jan Desktop

• Use an open-source desktop client with local and cloud provider choice.
• Serve a local model to another tool through an OpenAI-compatible endpoint.

What to weigh up

• Using a cloud provider changes the data boundary.
• Available model size and speed depend on hardware.

Models & access

Downloads local models and connects to cloud providers; exposes a local OpenAI-compatible server. Local operation can keep inference on-device.

Open-source client; inference separate Local hardware or chosen cloud provider supplies inference.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Jazz

• Running one assistant across interactive terminal work and scheduled scripts.
• Using local models for general assistant tasks.

What to weigh up

• Messaging and email integrations require their own setup.
• General-purpose assistant scope extends beyond coding; no quality ranking is inferred.

Models & access

Provider setup includes hosted APIs and local Ollama or llama.cpp.

Open-source client; selected model costs Model pricing depends on the chosen provider; local inference requires your own hardware.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Junie

• Use a coding agent within an existing JetBrains workflow.
• Use the CLI with a provider API key rather than a JetBrains model subscription.

What to weigh up

• The default model is dynamically chosen, not a stable model identifier.
• Verify IDE BYOK support in the installed plugin; official text currently conflicts.

Models & access

CLI supports BYOK and custom providers. The plugin models page is internally inconsistent about BYOK, so CLI capabilities must not be advertised as IDE parity.

JetBrains plan or CLI BYOK For CLI BYOK, usage is billed by the chosen provider and does not require a JetBrains AI subscription.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Kilo Code

• Use a common coding agent through VS Code, JetBrains or terminal.
• Switch providers/models with a gateway rather than changing the whole client.

What to weigh up

• Model catalogue breadth does not establish task quality.
• Autonomous command mode can approve actions and requires deliberate permission configuration.
• Custom subagent documentation specifically covers the CLI; interface parity is not assumed.

Models & access

Kilo Gateway, direct BYOK, model switching and local Ollama/LM Studio/Atomic Chat are documented.

Free client; optional inference, cloud compute and team plans Individuals can use free/local models or BYOK without a paid Kilo inference plan. Gateway usage, optional Kilo Pass credits and cloud compute are separate charges. Credit purchases carry a processing fee. Teams/Enterprise add collaboration and governance.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Kimchi

• Terminal coding with role-based model orchestration.
• Importing MCP configuration and skills from supported assistants.

What to weigh up

• Requires Kimchi service credentials.
• Remote-session teleport is explicitly unavailable on Windows in the README.

Models & access

CLI obtains its supported models from Kimchi metadata service and authenticates with a Kimchi API key. No arbitrary endpoint or local execution support established.

Kimchi service billing Kimchi model-service access is separate from the Apache-licensed client.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.

Why consider Kimi Code CLI

• Use Kimi with native terminal sessions, lifecycle hooks and ACP editors.
• Migrate existing legacy Kimi CLI data to the supported successor.

What to weigh up

• Windows requires Git Bash.
• Legacy Python Kimi CLI credentials, MCP authorizations and plugins do not all migrate automatically.

Models & access

Works with Kimi through OAuth or Moonshot API keys and can be configured for other compatible providers. No unconditional local-model guarantee established.

Kimi membership or provider API usage Kimi Code account access uses membership benefits with usage limits. Moonshot Open Platform API access is a separate pay-as-you-go route with different keys and endpoints. Other configured providers follow their own accounts and billing; the CLI software does not include unrestricted hosted inference.

Set up on your computer

Choose the operating system of the computer you want to install on. Browser tools may not need installation.