Aider alternatives: pick by why you are leaving

The nearest thing to a drop-in Aider alternative is cecli, an actively maintained Apache-2.0 fork that keeps Aider's exact bring-your-own-key route — the same two environment variables and the same mandatory openai/ model prefix — while adding the agent mode, MCP, skills, subagents and hooks Aider never shipped. If your reason for leaving is something other than staleness, the answer changes: a tool-using terminal agent points at OpenCode, Crush or Goose; delegated runs point at OpenHands; an editor workflow points at Cline, Kilo Code or Zoo Code; and a protocol other than chat completions points at Claude Code or Codex CLI.

One framing correction before the list. Aider is slowing, not dead. The repository is not archived, no maintainer has announced anything, and there is no deprecation notice anywhere. Everything below is a counted signal with a date attached, because that is all the evidence supports.

Why people are leaving Aider

The two official views of "latest release" disagree — PyPI serves 0.86.2 while GitHub Releases stops at v0.86.0 — so 0.86.1 and 0.86.2 shipped to PyPI without a GitHub release. Trust PyPI for what pip install gives you, and confirm in your own environment. Two open threads, #4751 on the project's future direction (opened January 7, 2026) and #4584 on the release gap (opened October 13, 2025), are users asking rather than maintainers answering. Treat them as community signal only.

Cadence is the soft reason. Three functional gaps are the concrete ones, and each maps to a different replacement:

1. The model table is frozen, and that changes how edits are made. In released 0.86.2, configure_model_settings tests model == ms.name — plain string equality — and all 40 openai/-prefixed rows in the shipped 313-entry table are OpenAI models, so a gateway id exact-matches only when it happens to reuse one of those OpenAI names. Anything else falls through to substring rules whose newest Claude entries are sonnet-4-5, opus-4-6 and haiku-4-5. An id like openai/claude-sonnet-4-6 matches none of them and inherits the dataclass defaults: edit_format "whole", use_repo_map False, cache_control False, no weak model and no editor model. Check your own id against those three substrings rather than assuming: openai/claude-haiku-4-5 contains haiku-4-5, so it takes the diff-edit branch with the repo map enabled, and only cache_control stays False. Read at aider/models.py and model-settings.yml on September 19, 2026. This is what the shipped code does, not something observed at runtime here.
2. No MCP. At tag v0.86.2 no file path in the repository contains mcp, and aider.chat's documentation index has zero occurrences of "MCP" or "Model Context Protocol".
3. One wire protocol. Aider's compatible-endpoint page documents exactly one third-party route: OPENAI_API_BASE, OPENAI_API_KEY and the openai/ prefix. Its Anthropic page names only the API key, and no base-URL override for Anthropic appears anywhere in the v0.86.2 source tree, so Aider documents no way to point its Anthropic route at another host. Undocumented is not the same as impossible — Aider calls models through LiteLLM, whose main.py reads ANTHROPIC_API_BASE from the environment — but that path is untested here and unsupported by Aider's own docs.

Which alternative matches which exit reason

Pick by the reason you are leaving, not by star count. Every project below was checked on September 19, 2026 and is under active development unless the row says otherwise.

Sources for that table, all read September 19, 2026: OpenCode providers, the Crush README, Goose providers, OpenHands LLM configuration, Cline's compatible-provider page, Claude Code environment variables and the Codex configuration reference. Kunavo's repository serves chat, messages and responses routes under /v1, so all three protocols have an endpoint — that is a protocol statement, not a compatibility test, and none of these clients has been runtime-tested against Kunavo.

Do not migrate to these

Four projects also changed identity in 2026, so setup instructions written earlier will send you to the wrong place: sst/opencode now redirects to anomalyco/opencode, block/goose to aaif-goose/goose after Block donated Goose to the Agentic AI Foundation on April 7, 2026, All-Hands-AI/OpenHands to OpenHands/OpenHands, and Anaconda acquired Kilo Code on July 15, 2026 — that announcement says Kilo "remains available today with no changes to existing products, plans, or support for current users" and defers integration details, so what changes later is unknown. Its site is now kilo.ai. Verify the redirect target before pasting an install command from an old post.

What migrates, and what you rewrite

cecli is the only option on this page where migration is a rename. Its compatible-endpoint page is Aider's with the product name and install command swapped: same two variables, same openai/ prefix, same instruction to pass a compatible endpoint. Its YAML config page names .cecli.conf.yml and its advanced model settings page names .cecli.model.settings.yml and .cecli.model.metadata.json — the same three files with a different prefix, loaded from the same home, git-root and current-directory locations.

# 1. Keep a working Aider install. Nothing below removes it.
cp .aider.conf.yml .aider.conf.yml.bak 2>/dev/null

# 2. Install the fork alongside it.
uv tool install cecli-dev

# 3. The same two variables Aider reads. Nothing is renamed here.
export OPENAI_API_BASE=https://api.kunavo.com/v1
export OPENAI_API_KEY=<your Kunavo key>

# 4. Rename the three dotfiles. cecli looks for the cecli-prefixed names.
mv .aider.conf.yml            .cecli.conf.yml
mv .aider.model.settings.yml  .cecli.model.settings.yml
mv .aider.model.metadata.json .cecli.model.metadata.json

# 5. Same prefix rule as Aider: openai/ is mandatory for a compatible endpoint.
cecli --model openai/claude-sonnet-4-6

# Rollback: rename the three files back and run aider again. Nothing else changed.

Two things are genuinely better in the fork rather than merely renamed. Its ModelSettings dataclass defaults edit_format to "diff" where Aider's defaults to "whole", so an unrecognised id starts from diff edits instead of whole-file rewrites — read at cecli/models.py on main, September 19, 2026. And it adds a first-class model-providers mechanism, where a custom OpenAI-compatible endpoint is declared with api_base, api_key_env and an optional models_url, instead of being squeezed through the single OPENAI_API_BASE slot. Its bundled table carries 383 entries against the released Aider's 313, including claude-sonnet-4-6 and claude-opus-4-7 — but it lists no 2026 Claude 5 id either, so a hand-written settings and metadata pair is still worth keeping. Two caveats stay on the record: its Anthropic page, like Aider's, documents only a key and no base-URL override, and use_repo_map still defaults to False.

Whatever you pick, migrate additively. Keep the old install and a copy of the dotfiles until one real task has completed on the new client, then delete the backups. The commands above never remove Aider, and the rollback is renaming three files.

What switching does to the bill

The client you choose changes token shape before it changes anything else, because the edit format decides how much the model re-emits. These are illustrative token arithmetic, not measured task costs and not a bill ceiling. Assume one session sending 200,000 uncached input tokens and receiving 12,000 output tokens under diff edits; for the whole-file column, assume the same session emits 40,000 output tokens because edited files are re-sent in full. That ratio is an assumption for illustration. Rates are live Kunavo catalog prices per million tokens.

Under these assumptions the edit format alone moves Claude Sonnet 4.6 from $0.312 to $0.480 for identical work. That gap is why a client's default for an id it does not recognise belongs in the comparison, and why it is not on its own a reason to switch: a settings file closes it without changing tools. Cheapest listed rate and lowest cost to finish the task are still different questions, and neither this page nor any price table answers the second one for your repository.

Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. Cache charges and external tools sit outside this example. The minimum top-up is $10 in prepaid credit — a funding minimum, not a task fee or a subscription. See billing details.

Where to start

If you are staying on the Aider-shaped workflow, nothing about your Kunavo configuration changes: the Aider setup guide documents the same OPENAI_API_BASE, OPENAI_API_KEY and openai/ prefix that cecli reads, and Aider pricing covers the settings file that restores diff edits, the repo map and the cost line. If you are changing client, start from that client's own page: OpenCode, Cline, Kilo Code, Claude Code or Codex CLI. These are published configuration references rather than compatibility tests. Create a Kunavo account when you are ready to fund a key, run one bounded task, and read the charge your account actually recorded for it.

Still choosing between agents rather than providers? OpenCode alternatives, OpenHands vs OpenCode and Goose vs Claude Code compare them head to head, and the OpenAI-compatible API guide covers what the shared protocol does and does not carry.

FAQ

Repository state, release dates, commit and issue counts, and every client configuration page linked above were read on September 19, 2026; Aider behaviour claims come from released tag v0.86.2 and cecli claims from its main branch. No client on this page has been runtime-tested against Kunavo, and no token bill was measured — every dollar figure is illustrative arithmetic over live catalog rates.