Agent AFK is an open-source AFK coding agent designed to run tasks while you are away. It plans, acts, verifies its own work, and notifies you when it finishes or needs input. Free, Apache-2.0, works with any model provider.

What does AFK mean for a coding agent?

AFK stands for Away From Keyboard. An AFK coding agent is designed to keep working after you walk away; it runs a task end to end, verifies the result, reaches a clear stopping point, and texts you when done.

How does Agent AFK notify me when it finishes?

Agent AFK sends a Telegram message when a run finishes, stalls, or needs your input. You can answer, redirect, or interrupt from your phone. Every run also writes a full trace to disk so you can see exactly what happened.

Is Agent AFK a good Aider replacement in 2026?

For most workflows, yes. Agent AFK covers everything Aider does as an interactive terminal tool and adds multi-agent orchestration, 23 skills, persistent cross-session memory, a cron daemon, and Telegram notifications that Aider does not have. Aider's genuine edge is git integration (/undo in particular) and its tree-sitter repo map. If those two features are central to your workflow, that is worth weighing. For everything else, Agent AFK is the more capable and more actively maintained option.

Is Aider still being actively developed?

As of October 2026, Aider's last stable release was v0.86.2 on February 12, 2026. No feature releases have shipped since. The main branch shows model metadata updates only. Community discussion in GitHub issue #4751 describes the project as having slowed significantly, with contributors noting it has become a one-person maintenance effort. Aider still works and its git integration and repo map remain valuable. But if active development and a growing feature set matter to you, that is a meaningful difference.

What does Agent AFK have that Aider does not?

Multi-agent orchestration with parallel sub-agents across isolated worktrees. 23 bundled skills with auto-routing (/mint, /review, /diagnose, /standup, and more). Cross-session SQLite memory with FTS5 full-text search. A built-in cron daemon for scheduled tasks. Telegram push notifications delivering a verdict with evidence. A full TUI with 52 slash commands, ghost-text, and tool-lane visualization. Named terminal states (Done/Blocked/Asking/Interrupted) with a runtime-enforced evidence requirement on Done. One-command migration from Claude Code.

What does Aider have that Agent AFK does not?

/undo: a command that rewinds the last AI edit and commit byte-for-byte with clean refusal semantics if the commit was already pushed. A tree-sitter repo map using PageRank over symbol references to inject ranked codebase context into every request automatically. A polyglot coding leaderboard that benchmarks models across six languages. LiteLLM-backed support for a broader set of model providers. These are real capabilities and the comparison is honest about them.

What is the architectural difference between the two tools?

Aider is a single-agent interactive session: one model turn at a time, one thread of execution, your terminal open throughout. Agent AFK is a multi-agent orchestration platform: sub-agents run in parallel across isolated worktrees, sessions produce a named terminal state with a required evidence pointer, and the daemon continues working while you are away from the machine. Agent AFK also operates fully interactively when you want that mode. Aider does not have a daemon mode.

Does Agent AFK have a repo map like Aider?

Not a built-in automatic one. Agent AFK sub-agents read files explicitly via tools and can parallelize reads across large codebases, but there is no tree-sitter-powered PageRank map injected into every request. For large repos where ambient codebase context matters as much as explicit file selection, Aider's repo map is a genuine advantage.

Does Agent AFK support the same models as Aider?

Agent AFK supports any provider that speaks the Anthropic or OpenAI API format: Claude, GPT-4o, Gemini, Mistral, DeepSeek, xAI, and local models via Ollama or LM Studio. Aider covers a broader set via LiteLLM and publishes a leaderboard to help you choose between them. Both tools let you bring your own key with no hosted inference required.

Yes. Agent AFK is open source under Apache 2.0 with no feature gates. You bring your own API key. Aider is also Apache 2.0 and free. Neither tool charges for the software itself. Your only cost is the API calls you make.

Comparison

Agent AFK vs. Aider

Aider showed what terminal AI coding could be: open source, model-agnostic, git-native, and fast. Agent AFK is what that idea became when you add multi-agent orchestration, persistent memory, 23 bundled skills, and a daemon that delivers a Telegram verdict while you sleep. The terminal stays. The ceiling moved.

At a glance

Pick AFK for: unattended runs — it runs while you are gone
Pick AFK for: multi-agent parallelism across isolated worktrees
Keep Aider for: battle-tested git integration and tree-sitter repo maps
Keep Aider for: community legacy and a broad open-source model leaderboard

Single-agent interactive tool vs. multi-agent orchestration platform

Aider is a single-agent interactive tool. You launch it, add files to the chat, watch diffs appear turn by turn, and stay present for the whole session. That design is not a flaw: it is the right model for interactive pair programming and it earned Aider nearly 50K GitHub stars over three years of active development. Agent AFK starts from a different premise. You are not there. A goal goes in, sub-agents fan out across isolated worktrees in parallel, verification loops run, and the session ends in one of four named terminal states: Done, Blocked, Asking, or Interrupted. Done must carry an evidence pointer (file path, commit SHA, test result) enforced by the runtime, not the prompt. The verdict lands on your phone. AFK also works fully interactively with a rich TUI, 52 slash commands, and ghost-text completion. It is an orchestration platform that also happens to be a great interactive terminal agent. Aider is the interactive tool that helped prove the terminal was the right place to start.

Feature comparison

Grounded in the actual implementations, not marketing copy.

Dimension	Aider	Agent AFK
Operating model	Single-agent interactive session. You direct each turn, watch diffs in real time, and stay present throughout.	Multi-agent orchestration platform. Sub-agents fan out in parallel across isolated worktrees. Also works fully interactively with a complete TUI.
How a session ends	You type /exit or Ctrl-D. Outcome is in your terminal scrollback.	One of four named terminal states: Done, Blocked, Asking, or Interrupted. Done requires an evidence pointer (file, commit SHA, test result) enforced by the runtime, not the prompt.
Development status	Last stable release: v0.86.2, February 12, 2026. No feature releases since. Community discussion (issue #4751) notes the project has slowed significantly. Still works, but not evolving.	Actively developed. Regular feature releases, open roadmap, and a growing skill library. The gap between the two tools is widening.
Notification and async handoff	Terminal output only. You watch the session live or check scrollback afterward.	Telegram push: the verdict lands on your phone. You can reply to an Asking state, approve, or interrupt from Telegram without opening a terminal.
Agent orchestration	Single agent. Architect mode splits planning from editing across two models in the same sequential session, not parallel sub-agents.	/diagnose forks independent hypotheses and races them across isolated worktrees. /review fans out simultaneous agents across security, correctness, API compat, and test coverage in one parallel wave.
Skills / automation library	No built-in skill system. Community workaround: the aider-skills PyPI package hacks skill injection via --read flags and /run commands.	23 shipped skills with auto-routing: /mint, /review, /diagnose, /standup, /document, and more. New skills installable via a single command.
Cross-session memory	No persistent cross-session memory. Repo map rebuilds cold on every launch.	SQLite store with FTS5 full-text search. Facts, decisions, and conventions survive context compaction and are queryable across sessions.
Git integration	Exceptional and battle-tested. Auto-commits every edit with a descriptive message, /undo rewinds the last AI commit cleanly (refuses if already pushed), /diff shows the cumulative session diff. Three years of community hardening.	Full git tooling via bash: commits, diffs, branch management, and worktree isolation for parallel sub-agents. No dedicated /undo shortcut; use standard git revert.
Codebase understanding	Repository map built with tree-sitter. Uses PageRank over symbol references to pack a compact, ranked snapshot into every request without sending whole files.	Sub-agents read files explicitly via tools and can parallelize reads across a large codebase. No automatic repo map on every request.
Scheduling	No built-in scheduler. Must be driven manually or via external cron.	Built-in cron daemon. Schedule recurring tasks (nightly standup, weekly review) with a single command. Verdict delivered via Telegram.
Interactive TUI	Clean, minimal terminal UI. Focused on the interactive pair-programming loop with good readline support.	Full TUI with 52 slash commands, ghost-text completion, tool-lane visualization, thinking-lane display, shell passthrough, and background sub-agent management.
Model support	Broad: Claude, OpenAI, DeepSeek, Gemini, Mistral, xAI, Ollama, LM Studio, and dozens more via LiteLLM. Publishes a polyglot coding leaderboard.	Anthropic and any OpenAI-compatible endpoint, including local models via Ollama or LM Studio. Per-task model routing supported.
Lint and test loop	--auto-lint and --auto-test are first-class flags. Feed failures back to the model and hold commits until they pass. Simple, transparent.	Test and lint loops run inside skills like /mint and /diagnose. The loop is part of the skill definition rather than a global flag.
License	Apache 2.0. Open source.	Apache 2.0. Open source.

Install freenpm i -g agent-afk · free, Apache-2.0

Where Agent AFK pulls ahead

Genuine architectural differences, not marketing claims.

It runs while you are gone

Aider requires you to stay present. The terminal must stay open, you direct each turn, and the session ends when you leave. Agent AFK is built for absence. State a goal, step away from the machine, and a Telegram message delivers the verdict when the work is done and verified. Overnight migrations, multi-hour refactors, and parallel investigations all run without a watchful developer in the loop.

Multi-agent parallelism across isolated worktrees

Aider processes one sequential turn at a time inside one session. Agent AFK fans out sub-agents across isolated git worktrees in parallel. /diagnose forks independent hypotheses and races them to a verdict. /review dispatches simultaneous agents across security, correctness, API compat, and test coverage. Work that would take Aider hours of sequential turns finishes in one parallel wave.

23 skills, shipping regularly

Agent AFK ships 23 bundled skills with auto-routing: /mint for test-driven feature development, /review for multi-dimensional code review, /diagnose for root-cause analysis, /standup for async status, /document for living docs, and more. Aider has no native skill system. The community aider-skills package exists, but it is a workaround that injects context via --read flags and /run commands rather than providing a native feature.

Persistent cross-session memory

Every project decision, architectural convention, and debugging breakthrough recorded in an Agent AFK session is stored in a SQLite store with FTS5 full-text search. That knowledge survives context compaction and is queryable in any future session. Aider rebuilds its repo map cold on every launch. The longer you use Agent AFK on a project, the smarter it gets.

Actively developed

Agent AFK ships regular feature releases. Aider last released v0.86.2 on February 12, 2026. Community discussion in issue #4751 notes the project has slowed significantly, with contributors describing it as effectively a one-person maintenance effort. Aider still works, but the distance between the two tools is growing with every AFK release.

Where Aider pulls ahead

Honest. Aider is a serious, well-regarded open-source project with real strengths.

Battle-tested git integration

Aider's /undo command rewinds the last AI edit and commit byte-for-byte, refusing cleanly if the commit has already been pushed. Auto-commit with conventional messages and /diff for cumulative session diffs. Three years of community hardening produced git semantics that no other tool has fully replicated. If your workflow centers on granular, reversible AI commits, Aider's git integration is still the benchmark.

Tree-sitter repo map with PageRank

Aider builds a compact, ranked snapshot of your entire codebase using tree-sitter symbol extraction and PageRank over cross-file references, then packs it into every request without sending whole files. This gives Aider strong ambient context on large repos without requiring explicit file selection. Agent AFK sub-agents read files explicitly; there is no equivalent automatic map.

Community legacy and model leaderboard

Nearly 50K GitHub stars, years of community-contributed tutorials, and a polyglot coding leaderboard that benchmarks every major model across six languages. The leaderboard is widely cited as a neutral reference even by developers who have moved on from Aider for daily use. That accumulated community knowledge is real and still valuable.

First-class lint and test loop

--auto-lint and --auto-test are built-in flags, not a skill to configure. Point Aider at your lint and test commands and it automatically feeds failures back to the model, holding commits until everything passes. The implementation is simple, transparent, and works with any test runner without additional setup.

Frequently asked questions

Is Agent AFK a good Aider replacement in 2026?: For most workflows, yes. Agent AFK covers everything Aider does as an interactive terminal tool and adds multi-agent orchestration, 23 skills, persistent cross-session memory, a cron daemon, and Telegram notifications that Aider does not have. Aider's genuine edge is git integration (/undo in particular) and its tree-sitter repo map. If those two features are central to your workflow, that is worth weighing. For everything else, Agent AFK is the more capable and more actively maintained option.
Is Aider still being actively developed?: As of October 2026, Aider's last stable release was v0.86.2 on February 12, 2026. No feature releases have shipped since. The main branch shows model metadata updates only. Community discussion in GitHub issue #4751 describes the project as having slowed significantly, with contributors noting it has become a one-person maintenance effort. Aider still works and its git integration and repo map remain valuable. But if active development and a growing feature set matter to you, that is a meaningful difference.
What does Agent AFK have that Aider does not?: Multi-agent orchestration with parallel sub-agents across isolated worktrees. 23 bundled skills with auto-routing (/mint, /review, /diagnose, /standup, and more). Cross-session SQLite memory with FTS5 full-text search. A built-in cron daemon for scheduled tasks. Telegram push notifications delivering a verdict with evidence. A full TUI with 52 slash commands, ghost-text, and tool-lane visualization. Named terminal states (Done/Blocked/Asking/Interrupted) with a runtime-enforced evidence requirement on Done. One-command migration from Claude Code.
What does Aider have that Agent AFK does not?: /undo: a command that rewinds the last AI edit and commit byte-for-byte with clean refusal semantics if the commit was already pushed. A tree-sitter repo map using PageRank over symbol references to inject ranked codebase context into every request automatically. A polyglot coding leaderboard that benchmarks models across six languages. LiteLLM-backed support for a broader set of model providers. These are real capabilities and the comparison is honest about them.
What is the architectural difference between the two tools?: Aider is a single-agent interactive session: one model turn at a time, one thread of execution, your terminal open throughout. Agent AFK is a multi-agent orchestration platform: sub-agents run in parallel across isolated worktrees, sessions produce a named terminal state with a required evidence pointer, and the daemon continues working while you are away from the machine. Agent AFK also operates fully interactively when you want that mode. Aider does not have a daemon mode.
Does Agent AFK have a repo map like Aider?: Not a built-in automatic one. Agent AFK sub-agents read files explicitly via tools and can parallelize reads across large codebases, but there is no tree-sitter-powered PageRank map injected into every request. For large repos where ambient codebase context matters as much as explicit file selection, Aider's repo map is a genuine advantage.
Does Agent AFK support the same models as Aider?: Agent AFK supports any provider that speaks the Anthropic or OpenAI API format: Claude, GPT-4o, Gemini, Mistral, DeepSeek, xAI, and local models via Ollama or LM Studio. Aider covers a broader set via LiteLLM and publishes a leaderboard to help you choose between them. Both tools let you bring your own key with no hosted inference required.
Is Agent AFK free?: Yes. Agent AFK is open source under Apache 2.0 with no feature gates. You bring your own API key. Aider is also Apache 2.0 and free. Neither tool charges for the software itself. Your only cost is the API calls you make.

Compare other tools

See how Agent AFK compares across the landscape.

Try Agent AFK

Open source, Apache-2.0, bring your own model.

Install free Read the docs View on GitHub