Aider vs Charm Crush (2026)

The agent works through git, runs configurable lint and test commands, offers to run shell commands it suggests, scrapes web pages into the chat with /web, and takes image and voice input. There is no MCP client, plugin system or integration surface for external services.

Sourceaider.chat/docs/config/options.html and usage/commands.html; aider.chat/docs/usage/commands.htmlread 2026-09-29

Crush connects to MCP servers over stdio, HTTP and SSE, uses language servers for diagnostics and references, fetches web pages, and runs the developer's own toolchain through its bash tool, alongside built in tools to view, grep, glob, edit and write files.

SourceCrush docs (configuration/mcp, tools overview) and github.com/charmbracelet/crushread 2026-09-29

Architect mode hands a change from an architect model, which proposes it, to a separate editor model, which turns it into file edits. After edits aider lints and optionally tests, then tries to fix the errors whenever a command returns a non zero exit code, a condition built into the flow.

Sourceaider.chat/docs/usage/modes.html and usage/lint-test.htmlread 2026-09-29

One agent works through a loop of file, search, language server and shell tools, with several sessions per project and models switchable mid session. There are no subagents, handoffs, branching or coordination across agents.

SourceCrush docs (tools overview, sessions) and github.com/charmbracelet/crushread 2026-09-29

Work starts from the person, through the interactive terminal chat, a one shot --message flag for scripts, and a watch mode that acts when the developer writes an AI! or AI? comment in any file. Nothing runs on a schedule, webhook or repository event.

Sourceaider.chat/docs/usage/watch.html and scripting.htmlread 2026-09-29

The crush run command is a non interactive mode for scripts and CI/CD pipelines. A published GitHub Actions workflow runs it on every pull request to review the diff, so work reaches the agent without a person starting it. The interactive terminal is the other surface.

Sourcegithub.com/charmbracelet/crushread 2026-09-29

A repository map lists the repo's files with the key symbols each defines and their critical definition lines. A graph ranking over a file dependency graph selects the most relevant parts to fit a token budget set by --map-tokens. Read only reference files and scraped web pages can be added as further context.

Sourceaider.chat/docs/repomap.htmlread 2026-09-29

Context is assembled per run from the repository through view, grep and glob tools, language server diagnostics and references, and project context files such as AGENTS.md and CRUSH.md. There is no index, graph or embeddings layer that persists and can be queried.

SourceCrush docs (context files, tools) and github.com/charmbracelet/crushread 2026-09-29

The chat is written to .aider.chat.history.md, and --restore-chat-history reloads the prior conversation on launch so work can continue, which is state the agent writes and carries across sessions. There is no stated lifetime or purge path beyond clearing the chat. Git commits record the code's changes and do not serve as memory.

Sourceaider.chat/docs/config/options.html and HISTORYread 2026-09-29

Sessions persist across restarts in a per project SQLite database and can be reopened, but the model remembers only the current session, and no context carries from one session to another. Project context files are instructions the developer writes.

Sourcegithub.com/charmbracelet/crushread 2026-09-29

Aider asks for confirmation before it runs a shell command it suggests, adds a file to the chat or fetches a URL. Bulk accept or reject covers a series of confirmations, and a --yes-always flag is off by default. In ask mode nothing is edited, and /undo reverts the last aider commit.

Sourceaider.chat/HISTORY, docs/faq.html and config/options.html; aider.chat/history.htmlread 2026-09-29

Crush asks the developer for permission before it runs tool calls, including in non interactive runs, with an allow list for trusted tools and a --yolo flag that skips the prompts only when the developer chooses it.

Sourcegithub.com/charmbracelet/crush and Crush docs (permissions, crush run)read 2026-09-29

As a local CLI, Aider has no hosted service, accounts or organization layer, so it offers and claims no SSO, role based access, audit logging or attestation. An .aiderignore file limits what enters the AI's context, and analytics can be switched off.

Sourceaider.chat/docs/config/options.htmlread 2026-09-29

Provider keys are supplied locally on the developer's own machine, where Crush runs. Usage metrics can be switched off, and prompts and responses are never collected. Ignore files limit what Crush reads, and Charm's optional Hyper inference service offers zero data retention. No certification, single sign on, role based access or audit logging is offered.

Sourcegithub.com/charmbracelet/crush, Crush docs (metrics and privacy) and hyper.charm.landread 2026-09-29

Every AI edit is committed to git with a generated message, so each change is a diffable record that /diff shows and /undo reverts. The session is written to a chat history file by default, with an option to log the full conversation with the LLM, and together these show what the agent changed and why.

Sourceaider.chat/docs/usage.html and config/options.html; aider.chat/docs/config/options.htmlread 2026-09-29

Crush writes a per project JSON log with timestamps and session IDs that records each tool call, such as a bash command, and each prompt sent and response received. The log can be viewed with crush logs, tailed or followed live, and debug mode adds language server traffic. Sessions and their history are kept in a local SQLite database.

SourceCrush docs (logging, sessions) and github.com/charmbracelet/crushread 2026-09-29

Aider is an Apache 2.0 CLI that runs on the developer's own machine and edits the local git repository. It works with local models through Ollama or any OpenAI compatible local server, so code and inference can stay on the customer's hardware.

Sourceaider.chat/docs/llms.html and github.com/Aider-AI/aiderread 2026-09-29

Code stays on the developer's machine because Crush runs locally in the terminal. It works air gapped once provider auto update is switched off and providers are updated manually, and can use local models through Ollama or LM Studio.

Sourcegithub.com/charmbracelet/crush and Crush docs (air gapped, local models)read 2026-09-29

An official Aider-AI conventions repository holds a community collection of ready made conventions files (Bash scripts, Flutter, functional programming, Go, Next.js with TypeScript and others), loaded with /read to steer the agent. These are guidance files, not prebuilt agents or task templates, and there is no further library.

Sourceaider.chat/docs/usage/conventions.html and github.com/Aider-AI/conventionsread 2026-09-29

Skills follow the open Agent Skills standard, with SKILL.md packages discovered from a skills directory. Crush ships no built in skills or catalog and points to Anthropic's public skills repository for examples, so customers assemble the skills themselves.

Sourcegithub.com/charmbracelet/crushread 2026-09-29

On the customer's own keys, Aider connects to OpenAI, Anthropic, Gemini, Groq, xAI, Azure, Cohere, DeepSeek, OpenRouter, GitHub Copilot, Vertex AI, Amazon Bedrock and any OpenAI compatible API, plus local models through Ollama and LM Studio. Main, editor and weak models are set separately, and /model switches mid session.

Sourceaider.chat/docs/llms.html and config/options.htmlread 2026-09-29

The developer picks the provider and model from 17 providers, including Anthropic, OpenAI, Google Gemini, AWS Bedrock, Azure, Groq, OpenRouter and Hugging Face, or uses custom OpenAI compatible or Anthropic compatible endpoints and local models. Models can be switched mid session while keeping the conversation.

SourceCrush docs (configuration/providers) and github.com/charmbracelet/crushread 2026-09-29

Scripting works from the command line with --message and from Python through the Coder class. The Python API is not officially supported or documented and may change without notice, and there is no HTTP API, SDK or MCP server.

Sourceaider.chat/docs/scripting.htmlread 2026-09-29

Fantasy, the Go agent framework Charm built to power Crush, is a library with a multi provider API and custom tools. Crush can be driven from scripts through crush run with stdin and stdout, and extended with Agent Skills. MCP support lets Crush reach other systems.

Sourcegithub.com/charmbracelet/fantasy and Crush docs (cli/run)read 2026-09-29

The repository ships a benchmark harness with setup and run instructions. It measures how well aider and a chosen model turn coding requests into code that passes unit tests, with custom models and edit formats and a YAML report of pass rates, cost and timing. The project publishes its polyglot leaderboard of 225 exercises in six languages from it.

Sourcegithub.com/Aider-AI/aider/benchmark and aider.chat/docs/leaderboards; github.com/aider-ai/aider/blob/main/benchmark/readme.mdread 2026-09-29

The agent runs project commands and tests through its bash tool and reads language server diagnostics, and debug logging exposes agent and language server behavior. These check the customer's code and help troubleshoot a run, but there is no harness to evaluate or regression test the agent's own behavior.

Sourcegithub.com/charmbracelet/crushread 2026-09-29

Aider edits files, runs shell commands and scrapes web pages into markdown with /web, but it does not drive a browser or operate software through a graphical interface.

Sourceaider.chat/docs/usage/commands.htmlread 2026-09-29

Everything runs through file tools, shell commands, language servers, MCP servers and a fetch tool that retrieves pages. Crush does not control a browser or operate software through a graphical interface.

Sourcegithub.com/charmbracelet/crushread 2026-09-29