lemoncrow

<!-- cspell:ignore Alamofire Excalidraw ast-grep codegraph ctags django jcodemunch nohit okhttp scip serena tokio vscode zoekt beasm Trendshift telegraphese --> <div align="center"> <img src="docs-site/favicon.png" width="36" height="36" alt="" style="vertical-align: middle;"> # LemonCrow Runtime ### Honest and benchmark proven — a faster Agent: one-shot code search, **37.7% fewer turns**, **23.7% less wall-clock** (and average 30% cheaper) Keep using Claude Code normally — LemonCrow sits underneath it and gives the agent better search, shorter file reads, compact command output, and reusable memory. [](LICENSE) [](https://github.com/lemoncrowhq/lemoncrow/releases) [](https://github.com/lemoncrowhq/lemoncrow) [](integrations/claude) [](integrations/codex) [](integrations/opencode) [](integrations/copilot) [](integrations/copilot-cli) [](integrations/cursor) [](scripts/install_hermes.sh) [](integrations/antigravity) **Live savings and time saved across LemonCrow sessions** [](https://lemoncrow.com) [](https://lemoncrow.com) [](https://lemoncrow.com) [](https://lemoncrow.com) [Install](#install-in-30-seconds) · [Check your savings first](#check-your-own-savings) · [Why trust the numbers?](#why-trust-the-numbers) · [Results](#results) · [Pricing](https://lemoncrow.com/pricing) [](https://lemoncrow.com/#terminal) *Click for the full walkthrough on [lemoncrow.com](https://lemoncrow.com/#terminal).* </div> --- ## Install in 30 seconds Run this once: curl -fsSL https://install.lemoncrow.com | bash Then create or sign in to a free LemonCrow account and activate it inside the project where you use Claude Code: cd your-project lc login lc init LemonCrow wires in better tools behind the scenes and starts tracking local savings as sessions finish. Remote telemetry is on by default; turn it off anytime with `lc telemetry remote off`. ```bash lc update ``` Check that everything is connected: ```bash lc doctor ``` ## What changes for you LemonCrow does not ask you to learn a new coding app. It improves the work Claude Code already does: | Before | With LemonCrow | | -------------------------------------------------------------- | -------------------------------------------------------------------------------- | | Claude greps, reads a whole file, greps again to find code. | One-shot code search returns the symbol, its callers, and exact ranges in a single call. | | The same context gets rediscovered again and again. | Useful session context can be reused, not re-searched. | | Long grep-and-read loops burn turns and wall-clock time. | Fewer round-trips: 37.7% fewer turns, 23.7% less wall-clock on SWE-bench Verified. | | Speed and savings are hard to see. | A local meter shows turns, tokens, and cost dropping in real time. | ### What actually gets replaced `lc init` gives Claude Code 5 tools and hides the built-ins behind them — one way to do each job, not two. **Find things in one shot** -- no wandering the codebase call after call. | LemonCrow tool | Replaces (hidden from the model) | Why | | --------------- | ---------------------------------- | ---------------------------------------------------------------------------------------------------------- | | `code_search` | Grep, Glob | One call returns the symbol, its callers/callees, and ranked source -- no grep-loop-then-read-whole-file. Ranked by call-graph centrality over a tree-sitter symbol table | | `read` | Read | Returns an outline or the exact`:L10-L40` range, budgeted, instead of the full file | | `edit` | Edit, Write | Verified, cross-file edits in one call instead of per-file patch-or-create guessing | | `bash` | Bash | Output is capped and structured so a noisy build log can't blow the context window | | `web_fetch` | WebFetch | Strips a page to clean Markdown instead of a raw HTML dump | What's unchanged: Claude Code itself, the model, your workflow. Full internals: [Architecture](docs/architecture.md). ### Why a runtime, not just tools A bare MCP server is a library the model can call *if* it remembers to. A runtime decides what's callable at all — four jobs, four layers: | Layer | Without it | With it | | -------------------------------- | -------------------------------------------------- | ---------------------------------------------------------------------------------------- | | **Agents** — process isolation | a "read-only" agent can still edit | `explore`/`plan`/`research`/`review` hard-deny `edit`/`write` at the host-config level | | **Skills** — standard library | multi-step procedures re-improvised every session | encoded once, invoked the same way every time | | **Hooks** — interrupts | wasteful re-reads; "done" without a check | bad calls denied before they run; on Claude Code, session close blocked until verification ran (advisory on other hosts today) | | **MCP tools** — syscall surface | agents fall back to grep-and-read under pressure | natives hidden — LemonCrow tools are the only surface for those jobs | ### The engine underneath — not just for code Under the coding runtime is a domain-neutral context engine with one contract: **one-shot retrieval** — a single call returns what the model needs, packed to a token budget; follow-up queries are allowed, retry loops are never required. Retrieval, compression, memory, and replay do not assume code — code is the first conforming retriever (`Retriever` protocol, `lemoncrow.core.capabilities.retrieval`), chosen because the pain is mature and the results are measurable. The same engine is being opened to teams building any LLM agent: [lemoncrow.com/engine](https://lemoncrow.com/engine). ### Agents Packaged in [integrations/agents/](integrations/agents/) — each a distinct capability grant (subagent name `lc:<mode>`), not a persona: | Agent | Writes? | Use | | ---------- | :-----: | ------------------------------------------------ | | `code` | Yes | default interactive — edits, refactors, features | | `auto` | Yes | fully autonomous — CI/headless runs | | `solve` | Yes | end-to-end solving of a well-defined task | | `execute` | Yes | one verified pass of an accepted plan | | `general` | Yes | catch-all for mixed work | | `bare` | Yes | minimal toolset, same discipline | | `explore` | No | read-only exploration — locate and cite | | `plan` | No | read-only planning, stops for human checkpoint | | `review` | No | adversarial read-only review | | `research` | No | external web research — cited memo | ### Skills Packaged in [integrations/skills/](integrations/skills/): | Skill | What it does | | ------------- | ------------------------------------------------------------ | | `lc` | manage LemonCrow itself via the CLI | | `benchmark` | measure savings on *your* repo — offline scan or live A/B | | `orchestrate` | one structured multi-step task, routed to the right surface | | `swarm` | N parallel attempts in isolated worktrees — best result wins | | `perf-review` | gate a change on measured performance, not read code | | `ux-review` | gate shipped UI on objective checks in a real browser | | `recall` | retrieve what past sessions learned | ## Check your own savings Do not take our 30% claim on faith — scan your own local agent history before installing: ```bash curl -fsSL https://savings.lemoncrow.com | bash ``` It reads local Claude/Codex session files, estimates where LemonCrow would have used fewer tokens or cheaper tool calls, and prints savings from your own history. Temporary local store — no account, no API keys. Useful variants: ```bash curl -fsSL https://savings.lemoncrow.com | bash -s -- --since 30d --top 10 curl -fsSL https://savings.lemoncrow.com | bash -s -- --host codex --limit 20 ``` ### Replay a past session `lc session replay` plays back a recorded session and, for each native call, **runs the real LemonCrow tool that would have replaced it** — grep/read loops collapse into one `code_search`, whole-file reads become budgeted outlines, `bash` logs get capped. It then estimates — from that session alone — the **cost, savings opportunity, and time** LemonCrow would have saved. No model re-run, nothing written; opens a shareable HTML page. ```bash lc session replay --last 1 ``` Works on Claude Code, Codex, and opencode sessions. The saving is an estimate; the live re-measured A/B is `lc benchmark local`. ## Why trust the numbers? A live badge alone proves little — anyone can fake a counter. Four checks instead: 1. **Raw benchmark receipts:** headline numbers link to committed per-task runs, costs, and reproduction commands. 2. **Your own scan:** the savings command checks your machine, not our marketing page. 3. **Labeled live badges:** aggregate usage counters — never the source of the 30% claim. 4. **Rows where LemonCrow does not win:** Terminal-Bench 2.1 is flat on accuracy (-0.2pp); it stays in the table. The trust is the audit trail, not the animation. ## Results Measured on the same model, same tasks, and same environment: | Benchmark | Baseline correct | LemonCrow correct | Correct delta | Baseline cost | LemonCrow cost | Cost delta | | ----------------------------------------------------- | -----------------: | ----------------: | --------------: | ---------------------: | --------------------: | -----------: | | SWE-bench Verified, 50 tasks x 5 reps | 80.8% | **92.8%** | **+12.0 pp** | $234.84 |**$165.45** | **29.5% cheaper** | | | SWE-bench Lite, 10 tasks x 3 reps | 93.3% | **100%** | **+6.7 pp** | $12.38 |**$10.79** | **12.9% cheaper** | | | SWE-bench Pro, 10 tasks x 5 reps | 88.0% | **90.0%** | **+2.0 pp** | $39.01 |**$30.61** | **21.5% cheaper** | | | Exploration tasks across 7 large repos x 5 reps | - | - | - | $19.11 |**$6.29** | **67% cheaper** | | | Telegraphic Q&A, 20 prompts x 5 reps | - | - | - | $8.93 |**$5.34** | **40.2% cheaper** | | | Terminal-Bench 2.1, 89 tasks vs public leaderboard* | 78.9% expected | 78.7% | -0.2 pp | $96.76 |**$69.52**† | **28.1% cheaper**† | | _{* LemonCrow 1 rep/task vs public leaderboard 5-rep average. † 5 timed-out tasks excluded from cost.} <p align="center"> <img src="benchmarks/cost_vs_savings_scatter.svg" alt="LemonCrow vs baseline: dollars saved per run against baseline task cost, across SWE-bench Verified/Lite/Pro, exploration, Telegraphic Q&A, and Terminal-Bench" width="720"> </p> SWE-bench Verified detail (250 runs a side) — one-shot search collapses the grep-and-read loop, so turns, wall-clock, and tool calls drop together: | Metric | Baseline | LemonCrow | Delta | | ---------------------- | ---------: | --------: | -----------------: | | Turns | 6,962 | 4,336 | **37.7% fewer** | | Wall-clock | 14.3h | 10.9h | **23.7% faster** | | Total tool calls | 6,700 | 4,167 | **-37.8%** | | Output tokens | 3.04M | 2.19M | **27.9% fewer** | | Bash | 3,327 | 1,785 | **-46.3%** | | Read | 1,733 | 1,050 | **-39.4%** | | Edit + Write | 1,628 | 759 | **-53.4%** | | Search (code_search) | - | 568 | lemoncrow-only | Exploration detail (7 large repos × 5 reps, read-only Q&A, no edits): | Tool | Baseline calls | LemonCrow calls | Delta | | ------------------------------- | ---------------: | --------------: | -------------: | | Read | 672 | 23 | **-96.6%** | | Bash | 508 | 71 | **-86.0%** | | Search (code_search) | - | 23 | lemoncrow-only | | Agent + orchestration calls\* | 79 | 1 | **-98.7%** | | Total tool calls | 1,259 | 118 | **-90.6%** | | input | 286,191 | 205,967 | **-28.0%** | | cache read | 35,862,919 | 2,753,393 | **-92.3%** | | cache write | 2,811,356 | 233,381 | **-91.7%** | | output | 426,367 | 68,893 | **-83.8%** | | input + cache write | 3,097,547 | 439,348 | **-85.8%** | Source: [`exploration_2026_06_29`](benchmarks/codebench/results/exploration_2026_06_29/) · [`telegraphic_2026_07_08`](benchmarks/codebench/results/telegraphic_2026_07_08). ## One-shot code search vs 10 named tools The search is the engine: right code in front of the agent on the first try. MRR and first-hit rate (rec@1) across ~7,200 query/gold pairs on 14 repos — 10 tools scored on the identical corpus. LemonCrow's rec@1 of 0.650 means the right code on the very first result two times in three, at 134ms p95: | Provider | MRR | rec@1 | p95 | | ------------------------------- | ----------: | ----------: | -------: | | **LemonCrow +semantic (BGE)** | **0.727** | **0.650** | 390ms | | LemonCrow lexical (default) | 0.676 | 0.582 | 134ms | | cocoindex-code (best rival) | 0.557 | 0.457 | 595ms | | serena | 0.401 | 0.359 | 3834ms | | ripgrep | 0.376 | 0.320 | 66ms | | universal-ctags (worst rival) | 0.237 | 0.226 | 1ms | No one had scored these 10 tools against each other on a shared query set before -- each publishes its own number, on its own terms, against its own baseline. ## Why it works Claude is strong; the loop around it is wasteful — grep, read a whole file, grep again. LemonCrow collapses that loop. - **One-shot search:** symbol, callers, and exact ranges in a single call — the biggest source of the turn and wall-clock savings. - **Better inputs:** exact file ranges, not whole files. - **Better outputs:** compact command output and replies, exact technical facts intact. - **Better memory:** context reused, not rediscovered. ## What you get - Works with Claude Code, Codex, Copilot, Copilot CLI, and opencode today; Cursor, Hermes Agent, and Antigravity integrations are in progress. Any MCP-compatible agent (LangChain, the OpenAI SDK, Gemini ADK, ...) can connect to the same tools. - Runs locally by default. - Source-available runtime: FSL-1.1-ALv2 engine, Apache-2.0 SDKs/integrations/docs. - Free account required to activate the official install. - Live local stats for cost, tokens, and savings; anonymous remote telemetry is on by default (opt out anytime). - Pro beta: existing gated features for $5/month or $49/year. ## Learn more - [Installation](docs/installation.md) - [Troubleshooting](docs/troubleshooting.md) - [Benchmarks](BENCHMARKS.md) · [full results, backed by docs](docs/benchmarks/results.md) · [every "vs" comparison, with sources](https://lemoncrow.com/vs) - [CLI reference](docs/cli.md) - [Architecture](docs/architecture.md) --- ## Star History <a href="https://www.star-history.com/?repos=lemoncrowhq%2Flemoncrow"> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/chart?repos=lemoncrowhq/lemoncrow&type=date&theme=dark&legend=top-left&sealed_token=Rn2b6rT3ghX0sl1jSS4wFIX77UahxINbyd-AgVtDAgV7BVA7aIINml5rE7v5bjzY82nrlzsmEASna5oYdS-JdGPZryfB2SRqgi8jNQY8VQl0Ra6W8QEVE6Bwn2Kd9bQzeEp03p3upVa48_1mbFJUhLQRp5lbXS8sEsNeQ_DK7_DfIRefJbXyjB27dHQN" /> <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/chart?repos=lemoncrowhq/lemoncrow&type=date&legend=top-left&sealed_token=Rn2b6rT3ghX0sl1jSS4wFIX77UahxINbyd-AgVtDAgV7BVA7aIINml5rE7v5bjzY82nrlzsmEASna5oYdS-JdGPZryfB2SRqgi8jNQY8VQl0Ra6W8QEVE6Bwn2Kd9bQzeEp03p3upVa48_1mbFJUhLQRp5lbXS8sEsNeQ_DK7_DfIRefJbXyjB27dHQN" /> <img alt="Star History Chart" src="https://api.star-history.com/chart?repos=lemoncrowhq/lemoncrow&type=date&legend=top-left&sealed_token=Rn2b6rT3ghX0sl1jSS4wFIX77UahxINbyd-AgVtDAgV7BVA7aIINml5rE7v5bjzY82nrlzsmEASna5oYdS-JdGPZryfB2SRqgi8jNQY8VQl0Ra6W8QEVE6Bwn2Kd9bQzeEp03p3upVa48_1mbFJUhLQRp5lbXS8sEsNeQ_DK7_DfIRefJbXyjB27dHQN" /> </picture> </a> --- ## License Open-core. The engine (`src/lemoncrow/core`, `bench`, `infra`, `gateway` minus `gateway/sdk`, plus `tests/`, `benchmarks/`) is licensed under the [Functional Source License, v1.1, ALv2 Future License](LICENSE) (FSL-1.1-ALv2) — free for internal use, non-commercial research, and professional services Converts to Apache-2.0 two years after each release. SDK bindings, host integrations, install scripts, and docs (`src/lemoncrow/sdk/`, `src/lemoncrow/gateway/sdk/`, `integrations/`, `scripts/`, `docs/`, `docs-site/`) are [Apache-2.0](LICENSE-APACHE).