What Is Aider? Terminal AI Pair Programming Explained: Features, Models, Settings and History

Aider is a free, open-source AI pair programming tool that runs in your terminal, edits the files in your local git repository and commits each change it makes. It is maintained by Aider AI LLC and released under the Apache 2.0 licence, and it works with almost any large language model you can reach through an API key, including models running on your own machine. This page explains what Aider is, what it costs to run, which models it supports, how its chat modes and git integration work, and which settings matter when a team adopts it. It is written for developers, engineering managers and the people who approve tools. Facts come from Aider's own website, documentation and GitHub repository. Last checked: 2026-11-15.

This is an independent guide from NHI Mgmt Group, based on Aider's public documentation as of November 2026. It is not sponsored or reviewed by Aider AI LLC. Product features change, so confirm settings against Aider's current documentation.

Key takeaways

• Open source and free to install. Aider is published on GitHub under the Apache 2.0 licence; the only running cost is the model provider you connect it to.
• Git is the safety net. By default Aider commits every edit with a generated message, commits your own uncommitted changes first, and offers /undo to discard its last change.
• Bring your own model. The docs cover cloud APIs, OpenAI-compatible endpoints, OpenRouter, cloud platforms such as Amazon Bedrock and Google Vertex AI, and local runtimes such as Ollama and LM Studio.
• Four chat modes. Code, ask, architect and help, with architect mode splitting planning and editing between two models.
• No hosted service or admin console. Control comes from configuration files, environment variables and command-line switches, so organisations govern Aider through the model accounts and repositories it touches.

What is Aider?

Aider describes itself as "AI pair programming in your terminal". You start it inside a project directory, add the files you want it to work on, and describe the change in plain language. Aider sends the request, the selected files and a compact map of the rest of the repository to the model you have chosen, then applies the edits the model returns directly to your files and commits them to git.

The project's first commit is dated 3 April 2023. The FAQ names Aider AI LLC as "the company behind the aider AI coding tool". The website's badges, as fetched on 15 November 2026, show about 44,000 GitHub stars, 6.8 million installs from PyPI and 15 billion tokens processed per week, and state that Aider wrote 88 per cent of the new code in its last release. These are the project's own figures.

Aider is a client, not a service. It runs on the developer's machine (or in a container), reads API keys from the environment or a .env file, and talks directly to the model provider. There is no Aider account, no Aider cloud and no Aider subscription.

Key facts at a glance

As of 15 November 2026, from Aider's website, documentation and repository.

Plans and pricing

Aider has no paid plans. The website and documentation list no subscription, seat price or enterprise tier, and the software is free to use, modify and redistribute under Apache 2.0. What you pay for is model usage.

Keeping costs visible: the /tokens command reports how many tokens the current chat context uses, and --cache-prompts turns on prompt caching for providers that support it, which the docs say covers the system prompt, read-only files, the repository map and the editable files. Because spend sits with the model provider, budget limits and alerts are set in the provider's console, not in Aider.

Models available

Aider uses the LiteLLM library to reach providers, so the list of models is long and changes with each provider. The documentation has dedicated connection pages for the following, listed alphabetically.

• Main, weak and editor models: Aider can use three roles. The main model does the work, --weak-model handles cheaper jobs such as writing commit messages and summarising history, and --editor-model turns an architect model's plan into file edits.
• Recent additions: the unreleased main branch notes in the release history add Claude 4.5, 4.6 and 4.7 aliases, Gemini 3 preview models and GPT-5.1 to GPT-5.4 variants. Until a new release is tagged, these apply to installs from the main branch.
• Capability warning: the docs say Aider "may not work well with less capable models", and point to Aider's own LLM leaderboards for code editing results.

Features and modes

The docs recommend moving between ask and code modes: discuss the approach in ask mode, then switch to code mode so the agreed plan is carried into the edit. Files can be added as editable (/add) or read-only (/read-only or --read), which is the usual way to supply coding conventions or reference files without letting Aider change them.

Supported IDEs and platforms

• Terminal first: Aider runs on macOS, Linux and Windows. The aider-install helper and the uv-based one-line installers set up Aider in its own Python environment, installing Python 3.12 if needed.
• Any editor through watch mode: with --watch-files, Aider responds to AI comments saved from whichever IDE or text editor you use, so there is no editor plugin to install.
• Browser UI: aider --browser launches a browser version of Aider.
• Containers: the docs publish Docker images (paulgauthier/aider and paulgauthier/aider-full) and show running them with the project directory mounted and the user ID mapped.
• Hosted dev environments: install pages cover GitHub Codespaces and Replit.
• Languages: the docs state Aider works with over 100 programming languages, with repository map support for those that have tree-sitter tags.

Enterprise and admin controls

Aider has no organisation account, SSO, SCIM, admin console or central audit log, because there is no hosted service to attach them to. Control is applied in three places: the configuration Aider reads, the model provider accounts it uses, and the repositories it commits to.

What this means for a team: a shared .aider.conf.yml committed at the repository root is the closest thing to a team policy, and a user's home or local file can override it. Pin model choice, turn on --git-commit-verify so existing pre-commit secret scanning and linting still run, and rely on branch protection and pull request review for anything that reaches a shared branch. The API keys Aider uses are long-lived machine credentials and should be scoped and owned like any other non-human identity; see AI Coding Tools and Secrets Sprawl.

Data handling and privacy

• Where code goes: Aider sends prompts, the files you add, read-only files and the repository map to the model provider you configure. Retention and training terms are therefore the provider's, so review them for each provider and account type. With a local model, code stays on your hardware.
• Analytics: the analytics page says collection is opt-in and that Aider "never collects your code, chat messages, keys or personal info". When enabled it records which models and edit formats are used, token counts, feature and command usage, and errors, tied to a random UUID4 identifier.
• Random opt-in prompt: if neither --analytics nor --no-analytics is set, Aider may ask a random subset of users to opt in. Running aider --analytics-disable once stops this permanently.
• Privacy policy: Aider AI LLC's privacy policy covers its website and tools, including device and usage information and the analytics data above.
• Commit messages: the weak model receives a copy of the diffs and chat history to write each commit message, so that model's provider also sees the change.
• Secrets on disk: keys can live in .env, .aider.conf.yml or the environment. Keep these files out of version control.

Release history

Milestone versions with the date of the tagged commit in the GitHub repository. The release history page lists changes per version without dates.

Limits and known constraints

• Model quality decides results: the docs warn that less capable models may return code Aider cannot apply to your files.
• Git expected: Aider "works best with code that is part of a git repo". --no-git is possible, and the docs then advise keeping your own backups.
• Single developer tool: there is no shared workspace, central policy, usage reporting or audit trail beyond git history and provider logs.
• Context limits: the model's context window caps how many files and how large a repository map fit in one request; adding only the files that need changing is the documented approach.
• Release cadence: the last release tag is February 2026 and newer model support sits on the main branch, so check which version is installed before relying on a model alias.
• Prompt caching: support depends on the provider, and the docs note caching statistics and costs are not always reported.

Security guidance

Aider runs with the permissions of the user who starts it and edits real files, so the practical risks are the ones common to terminal coding agents: untrusted content in the repository steering the model, shell commands proposed by the model, and API keys stored in plain files. The wider discipline is covered in the Open-Source Coding Agents Security Guide and the AI IDE and Coding Agent Security Guide. For Aider specifically:

• Keep confirmations on: avoid --yes-always on developer machines, and read each proposed shell command before approving it.
• Run hooks: set git-commit-verify: true so pre-commit secret scanning still runs on Aider's commits.
• Limit scope: use .aiderignore and --subtree-only to keep secrets, infrastructure code and unrelated services out of the chat.
• Treat repository text as input: files, read-only references, scraped web pages and AI comments all reach the model; review diffs before pushing. See Coding Agent Configuration File Attacks.
• Isolate unattended runs: scripted or CI use belongs in a container with a scoped key and no access to production credentials.

Practitioner checklist

• Decide which model providers and account types are approved, and review each provider's retention and training terms.
• Issue each developer or pipeline its own scoped, budget-capped API key with a named owner and rotation date.
• Commit a team .aider.conf.yml that pins the model, sets git-commit-verify: true and leaves yes-always off.
• Add .env and any local Aider config holding keys to .gitignore, and scan for committed keys.
• Create an .aiderignore for secrets, credentials and generated files in each repository.
• Run aider --analytics-disable on managed machines if your policy requires no telemetry.
• Protect shared branches and require pull request review for commits Aider makes.
• Record the installed Aider version and recheck the release history before upgrading.

Standards and references

• Aider: AI pair programming in your terminal (home page)
• Aider AI: Aider GitHub repository
• Aider AI: Apache License 2.0 (LICENSE.txt)
• Aider: Release history
• Aider: Git integration
• Aider: Options reference
• Aider: Connecting to LLMs
• Aider: Analytics

Related NHI Mgmt Group resources: Open-Source Coding Agents Security Guide · AI IDE and Coding Agent Security Guide · AI Coding Tools and Secrets Sprawl · AI Coding Assistant Security Checklist