2B

A local-first coding agent for the terminal. It runs your own models over Ollama's native protocol, keeps them focused instead of hallucinating, and gives you a full-screen TUI — streaming replies, a live plan checklist, narrated tool actions — without ever routing your local model through a translation layer that would confuse it.

I built 2B to keep working when the power and the internet don't — I live somewhere the grid isn't a guarantee, and I wanted a coding agent that doesn't fall apart the moment I'm offline.

macOS only. 2B is built and tested for macOS — the installer is a shell script that leans on Homebrew, and the clipboard integration uses pbcopy. It hasn't been tested elsewhere.

Why I built it

I kept being told that small models — Nemotron 3 Nano 4B, gpt-oss:20b, the Qwen family — were "good enough" for agentic coding. On paper they were. In practice, every off-the-shelf harness I tried broke them:

• opencode made gpt-oss invent tool names that didn't exist and emit fake <command> tags as plain text.
• Cline, Goose, Continue, OpenHands each failed in their own way — malformed tool schemas, reasoning collapse, the model narrating tool calls instead of making them.

The models weren't the problem. The harnesses were. Nearly all of them talk to a local model through a generic OpenAI-compatible /v1 shim and pile abstraction on top of it. That shim measurably degrades a small model's tool selection, and the extra complexity buries whatever capability the model actually has.

The one thing that worked cleanly was a ~350-line script I wrote that talked to Ollama's native /api/chat endpoint with a tiny, fixed set of five tools. So I grew that prototype into a real, shareable tool. That's 2B.

The core rule, and the whole point: all complexity lives on the host side. The model's world never changes — the same five tools, the same native wire format for whatever provider is active, no generic shim. Everything you see below — the TUI, the plan checklist, task management, model switching, auto-compaction — is something 2B renders around the tool loop, never a new thing the model has to understand.

What it does

The one idea that makes it work: the model only ever sees a small, fixed set of file-oriented tools, communicated in a way that stays close to each provider's own conventions rather than a one-size-fits-all shim. Most of what follows is functionality that 2B builds around that small surface, so a smaller model's world stays simple while the experience on screen is a full coding agent.

Reliable with small models — the host absorbs most of the difficulty:

• Edits that survive drift. Edits are matched using a layered fallback approach, so a near-miss can still land without falling back to something too uncertain. The interface the model works with doesn't change; the extra tolerance lives entirely on the host side.
• Catches its own mistakes. After a change, 2B runs some form of project-appropriate validation and works any new problems back into the same exchange — so the model becomes aware of issues it introduced without needing a separate mechanism for it.
• Finds definitions, not just matches. Searches are weighted to prioritize where something is actually defined, and file reads can include a structural summary — using deeper language tooling when it's available, and a simpler fallback approach otherwise.
• Rescues weak tool-calls. The system can recognize and recover intended actions that a weaker model expresses imperfectly, and nudges things along when a model describes an action without quite completing it.

A fuller experience, not a log dump:

• A live, continuously updating terminal view that shows overall progress and rough resource usage, including some indication of how full the working context is.

• Narrated actions — steps described in plain language with a simple success/failure tree, rather than raw tool output:

  ├ ✓ Searching for "MemoryScopeLevel" in lib
  └ ✓ Editing lib/memory/memory_store.dart
  

• Visual customization, reliable copy behavior across environments, keyboard-driven navigation, and support for running tasks in the background while you keep working.

Many models, one conversation. A broad range of providers — local and cloud — are supported, each communicated with in a way suited to that provider. The conversation history isn't tied to any one provider, so you can move between models mid-task without losing what's been established: start small and local, and move to something larger only if you need to.

Keeps the thread — or doesn't, your call. Depending on the setup, context can either persist automatically across messages or stay minimal until you choose otherwise. You can always start clean, and a full session — including the underlying actions taken — can be exported for reference.

Knows your project. A compact project summary can be generated and kept loaded automatically, with an on-demand outline view available too — all kept within limits so it won't overwhelm a smaller model's context.

Runs things, safely. What a model is allowed to execute depends on where it's running: more constrained locally, broader (but sandboxed) in cloud contexts, with writes confined to the project workspace. Riskier actions get an extra confirmation step regardless of settings, sensitive environment values are kept away from anything executed, and outputs from tools are treated as untrusted input so they can't be used to manipulate the model.

Scales without bloating the model. Larger or cloud-capable models can offload certain investigative or editing work so it doesn't consume the main conversation's space; external tool integrations are opt-in individually; older parts of a long conversation get automatically summarized as space runs low; and sessions are saved so they can be resumed later. Recent changes can be undone.

A few distinct modes, switchable on demand: one that confirms before writing, one that applies changes automatically, and one that only investigates and proposes without making any changes.

Install

One line — paste it in your terminal:

curl -fsSL https://raw.githubusercontent.com/dea6cat/2b-agent/main/install.sh | sh

It installs uv if you don't have it, installs the 2b command, then — on an interactive terminal — walks you through local model setup:

1. Optional clean install — offers to remove other agentic tools that proved unreliable with local models (opencode, Continue, Goose, Cline, OpenHands) and their configs. Off by default; it asks first.
2. Grades your machine — reads your RAM and chip and rates each candidate model (✓ fits well / ~ tight / ✗ needs NGB+), defaulting to the best one your hardware can run.
3. You pick one or several from the menu.
4. Installs Ollama and pulls what you chose, with a live progress bar.
5. Self-tests each model — tok/s + GPU residency, then a correctness check that runs a real one-line edit through 2B itself and verifies the result (✓ correct / ✗ wrong, ~20–90s per model). It only reports — it never removes a model — and --no-benchmark skips it. Then it prints how to launch 2B.

Already have Ollama and some models? It skips what you already have — it lists your installed models, offers to just use them (pulling nothing), and marks anything in the menu you've already got. Your existing setup is left untouched.

Prefer to do it by hand? Install the published package from PyPI, then run the same onboarding the installer uses:

pip install 2b-agent      # or: uv tool install 2b-agent
2b setup                  # grades your machine, installs Ollama, pulls a model, self-tests, fixes PATH

On macOS/Linux you can also use Homebrew (it puts 2b on your PATH automatically):

brew install dea6cat/2b/twob-agent   # the formula is twob-agent; it installs the `2b` command
2b setup

2b setup is the single source of truth for onboarding — the curl … | sh installer just installs uv + the 2b command and then runs it, so you get the exact same setup either way. (On first launch with no model configured, 2b offers to run it for you.)

The installer — and 2b setup — are scriptable: --yes (accept defaults, no prompts), --clean / --no-clean, --models "qwen3.5:9b qwen3:8b", --no-models, --no-benchmark (skip the correctness check), --fix-path / --no-fix-path (add uv's tool dir to your PATH for you via uv tool update-shell, or leave it — otherwise it asks, and never edits a profile without consent). Pass them through the pipe with ... | sh -s -- --yes --models "qwen3.5:9b".

Use it

2b                       # start in the current directory, autodetects a local model
2b "add a docstring to lib/main.dart"   # run one task, then drop into the session
2b --model qwen3.5:9b    # pin a model
2b --continue            # resume your most recent session (--resume <id> for a specific one)
2b --list-sessions       # list saved sessions
2b --list-models         # what's available across configured providers
2b --doctor              # diagnose PATH, Ollama, and the default model, then exit
2b --test                # grade installed models + compare them to the latest coding models
2b --test auto           # auto-clean: remove failures, then pull/coding-test the best new one
2b --update              # upgrade to the latest release (uv tool upgrade)
2b --rm                  # uninstall 2B and delete its config (asks first); --rm --yes to skip

Then just type what you want done. Type / to see the commands.

Testing your models

2b --test grades each installed model — tok/s, GPU residency, and a real two-change edit run through 2B (✓/✗, up to ~2 min each) — then prints a KEEP/REMOVE table with a suggested default. It also compares your models to the latest tool-capable coding models on ollama.com that fit your RAM, surfacing families you don't have and larger variants worth upgrading to. Plain --test only reports — it changes nothing (2b --test <model> grades just one).

2b --test auto is the hands-off cleanup:

• Removes the failing models automatically — no prompt (that's the point of auto); your current default is never removed.
• Then offers to pull + coding-test the best new candidate. It asks once before the multi-GB download (skip that with --yes), keeps the model only if it passes the coding test, and removes it if it fails — remembering a failed one so it isn't re-downloaded on the next run.

Updating

One command, whatever you installed with — it detects the method and runs the right upgrade:

2b --update

That resolves to uv tool upgrade 2b-agent (the curl … | sh installer / uv), pipx upgrade 2b-agent (pipx), or pip install -U 2b-agent (pip). You can of course run the matching command yourself — e.g. if you installed with pip:

pip install -U 2b-agent

2B also checks for a newer release in the background (at most once a day, never blocking startup) and prints a one-line notice on the next launch when one is available — set TWOB_NO_UPDATE_CHECK=1 to turn that off. Releases are tagged vMAJOR.MINOR.PATCH.

Providers

Local Ollama needs nothing. For anything else, set the matching environment variable and it shows up automatically in /models:

Or connect one from inside 2B — /connect <provider> prompts for the key with a hidden field and saves it to ~/.config/2b/keys.json (chmod 600) so it's there next time; /connect on its own shows what's connected, and /disconnect <provider> removes a saved key. A key exported in your shell always takes precedence over a saved one.

Switch models anytime with /model <name>. A bare name works when it's unambiguous; otherwise use provider:model (e.g. /model anthropic:claude-sonnet-5).

Commands

Keyboard

MCP servers (extra tools)

2B can pull in tools from MCP servers (stdio) like dart or mempalace. But its whole reason for existing is that small local models break when you flood them with tools — so MCP tools are opt-in and curated per tool: you enable a server and pick exactly which of its tools reach the model. Nothing is exposed until you say so.

Declare servers the usual way — a Claude-Code-style mcpServers block in ~/.config/2b/mcp.json (or ./.mcp.json in a project, which wins):

{
  "mcpServers": {
    "dart": { "command": "dart", "args": ["mcp-server"] }
  }
}

Then curate from inside 2B:

/mcp                          # servers and how many tools each has enabled
/mcp tools dart               # list a server's tools ([x] = enabled)
/mcp enable dart hot_reload analyze_files
/mcp enable dart all          # expose everything (careful on small models)
/mcp disable dart hot_reload  # or /mcp disable dart  to turn the whole server off

Enabled tools appear to the model as server__tool (e.g. dart__hot_reload) and their results come straight back into the loop. Only the tools you enable are ever sent — the model's tool list stays as small as you keep it.

Configuration

• Context window (local) — sized to your machine. For a local model 2B works out the largest window your box can run comfortably and pins num_ctx to it on every request (Ollama otherwise defaults to ~4k regardless of the model). It reads the model's architecture and trained max from /api/show, computes the KV-cache cost per token, and fits it into the RAM left after the model weights plus a headroom reserve — so a 16 GB laptop, an 18 GB one, and a 64 GB workstation each get a different, appropriate window (e.g. qwen3.5:9b ≈ 13k on 18 GB), never more than the model was trained for. That number drives auto-compaction (~75%) and the read-a-section threshold. Set TWOB_CONTEXT_TOKENS to override (higher if you want to spend more RAM, lower to save it).

• Verify loop. After a turn lands edits, 2B runs the project's own checks (analyze/lint/typecheck, test suites — whatever it detects for your stack) as local subprocesses — write-confined under the same workspace sandbox as run_command — and on failure feeds the errors back for a bounded fix loop (max 2 rounds). Nothing is sent anywhere.

• Context retrieval. On a new task (fresh conversation only, not a continued thread), 2B builds a short, host-side list of the files most likely relevant to your request — ranked by how they reference each other and the terms in your task — and injects it as a pointer mini-map to help the model orient. It's dropped when confidence is low, and it never replaces the model reading the files itself.

• Live reasoning. For thinking-capable models (Ollama thinking models like qwen3/gpt-oss, and Gemini 2.5), 2B streams the model's reasoning live, dimmed, as it arrives, then collapses it to a one-line 💭 thought for Ns once the reply starts. Controlled by /think.

• Environment toggles. Everything on-by-default can be turned off: TWOB_CONTEXT_TOKENS (override the local window) · TWOB_NO_DIAGNOSTICS (skip post-edit checks) · TWOB_NO_LSP (regex symbol map instead of a language server) · TWOB_NO_SEATBELT / TWOB_SEATBELT=strict (relax / harden the run_command sandbox) · TWOB_NO_TRIM (keep bulky tool output in each request) · TWOB_NO_HISTORY (don't persist sessions) · TWOB_SUBAGENT_MODEL (run delegated sub-agents on a cheaper model) · TWOB_NO_UPDATE_CHECK (no background update check) · TWOB_NO_VERIFY (turn off the verify loop) · TWOB_VERIFY_FAST (skip test suites, static checks only) · TWOB_VERIFY_CMD="cmd1;;cmd2" (declare your own checks for stacks 2B can't auto-detect) · TWOB_THINK=off|on|low|medium|high (persistent reasoning level — the /think default; applies to reasoning-capable Ollama models and Gemini 2.5; Anthropic and OpenAI-compatible providers not yet) · TWOB_NO_THINK_DISPLAY (hide the live reasoning stream — the model still thinks, just silently) · TWOB_NO_RETRIEVAL (skip the context retrieval pointer block) · TWOB_RETRIEVAL_FILES=N (how many files it lists, default 8).

Honest caveats

• It reads and writes wherever you point it — with guardrails. 2B resolves absolute paths and paths outside the working directory on purpose; it's a personal tool for your machine. Writes are still confirmed in normal mode, plan mode refuses them entirely, reads of known-secret paths (~/.ssh, ~/.aws, …) prompt first, and on the cloud path run_command's writes are sandbox-confined to the workspace by default (see the sandbox bullet above). The command classifier is defense-in-depth, not an unbypassable boundary — the sandbox is the boundary.
• Switching to a stronger model mid-task hands it a tool-call history it didn't make. For these five simple tools that's low-risk (the wire format is unambiguously the new provider's own; only the choices inside came from a weaker model), but you may see mild "why did I read that file" moments. It is not the shim-degradation failure that sank the other harnesses.
• It's a full-screen TUI. That means it lives in the terminal's alternate screen, so a single mouse drag selects what's on screen, not scrolled-off history. For a classic inline REPL, run 2b --classic.

License & privacy

2B is source-available under the PolyForm Noncommercial License 1.0.0 — free for noncommercial use; commercial use needs a separate license (see LICENSE, or open an issue). On first run you're asked to accept it once.

2B is local-first and has no telemetry. With a local model your code stays on your machine; with a cloud provider your prompts and the files the agent reads are sent to that provider under their terms. Full details — what's stored, where, and the opt-out environment variables — are in PRIVACY.md.

Built for local models, kept on task.