Semble
Semble gives AI coding agents exact code snippets via local code search — ~1 ms queries, ~99% fewer tokens than grep+read.
If your agent burns context on grep-and-read, Semble is the cheapest fix available: MIT-licensed, runs on CPU with no API keys, drops to ~1 ms queries against a ~500 ms index build. The reranking layer — definition boosts, identifier stems, noise penalties — is what separates it from naive embedding search on code. Treat the vendor's NDCG@10 of 0.854 as a starting point to verify on your own repo, not a guarantee.
Verified 20h ago · liveness 69/100 · cite: rightaichoice.com/tools/semble
- Developers running AI coding agents against large repos where grep+read exhausts context
- Teams that need local, private code search with no cloud calls or API keys
- Agent builders adding code retrieval to MCP workflows
- Solo devs trying to cut token cost and query latency on big monorepos
- Users who want a hosted code search service with a web dashboard
- Teams needing role-based access control, audit logging, or search analytics
- Debugging test helpers or .d.ts type stubs, which the default noise penalties down-rank
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Semble if you need a hosted code search service with a web dashboard, role-based access control, or audit logging — Semble is a local library you install into your agent, not a managed platform.
Semble itself is MIT-licensed at $0, so the cost that bites is your setup time: you install uv, run the interactive installer, and choose MCP, AGENTS.md, or a sub-agent integration per machine.
Semble costs nothing: it's MIT-licensed open source, and the published tier is $0. That makes price comparisons against hosted code-search services straightforward — those bill per seat or per query, while Semble's cost is the engineering time to install it and the CPU you already have. It fits solo devs and small teams running agents on large private repos; it is not aimed at buyers who want to purchase a managed service with SLAs, dashboards, and support contracts.
In short
Semble — Semble gives AI coding agents exact code snippets via local code search — ~1 ms queries, ~99% fewer tokens than grep+read. Best for Developers running AI coding agents against large repos where grep+read exhausts context, Teams that need local, private code search with no cloud calls or API keys, Agent builders adding code retrieval to MCP workflows. Free to use.
What's new in Semble
Checked 3 days agoAcross the latest 2 updates: 1 changelog entry and 1 news mention.
What people actually say about Semble — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
41 mentions across 4 sources (Hacker News, Product Hunt, GitHub, Lemmy) · researched Jul 3, 2026.
Average across the 4 sources that answered — each source counts once, not each post.
- +Near-instant indexing: 250ms for an average repo, faster than alternatives.
- +Query latency of ~1.5ms — snappy even on large codebases.
- +Claims 98% fewer tokens than grep+read, with verifiable savings per repo.
- +Zero-setup: runs on CPU, no API keys, no GPU required.
- +Natural-language code search with semantic + lexical hybrid retrieval.
- −MCP integration hangs on Codex-cli and other agents, causing frustration.
- −First-run session can deadlock on Windows 10 for up to 20 minutes.
- −No agent-level benchmarks to support the 98% token savings claim.
- −Custom tree-sitter grammars not supported, limiting language coverage.
- −Requires arcane instructions to coerce Claude Code into compliance.
- • No hidden costs, but token consumption from storing index on disk may affect disk space on very large repos
Viability Score
How well maintained and how widely used is Semble? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Natural-language code search that returns only relevant chunks to agents
- Code-aware file chunking via tree-sitter
- Static Model2Vec embeddings using potion-code-16M-v2
- BM25 lexical retrieval for identifiers and API names
- Reciprocal Rank Fusion merging semantic and lexical scores
- Adaptive weighting for symbol-like vs. natural-language queries
- Definition boosts so defining chunks outrank referencing chunks
- Identifier stem matching (query 'parse config' boosts parseConfig)
- File coherence reranking across chunks from the same file
- Noise penalties for test files, legacy shims, examples, and .d.ts stubs
- Search a local path, an https:// git URL, or several repos together
- MCP server exposing 'search' and 'find_related' tools
- Interactive 'semble install' that detects installed coding agents
- Content-type filtering: code, docs, config, or all
- Runs on CPU with no API keys, GPU, or external services
About Semble
Semble is an MIT-licensed, open-source code search library that plugs into AI coding agents so they stop grepping whole repositories and reading entire files into context. Your agent asks a question in plain language — "How is authentication handled?" — and gets back only the code chunks that answer it. It accepts a local path, an https:// git URL, or several repos at once to search them together. Retrieval is built for source code specifically. Files are split into tree-sitter chunks, then scored twice: static Model2Vec embeddings using the code-specialized potion-code-16M-v2 model for semantic similarity, and BM25 for exact matches on identifiers and API names. Reciprocal Rank Fusion merges the two lists, then code-aware reranking takes over — definition boosts so the chunk that defines a symbol beats chunks that just reference it, identifier stemming so "parse config" lifts parseConfig and config_parser, file coherence across matching chunks, and noise penalties that demote test files, legacy shims, examples, and .d.ts stubs. Setup is one interactive command: `semble install` detects your installed agents (Claude Code, Cursor, Codex, Gemini CLI, OpenCode) and lets you enable an MCP server, AGENTS.md CLI instructions, or a dedicated sub-agent. Two MCP tools ship — search and find_related. Indexes cache to disk on first search and reindex incrementally as files change. Semble is for developers running coding agents against large repos who care about token spend and keeping source on their own machine. It's a library you install, not a managed platform — there's no dashboard, permission model, or team analytics layer.
Behind the Verdict
We'd reach for Semble the moment an agent session on a monorepo starts dying by a thousand file reads. The pitch is narrow and honest: instead of the agent grepping, then opening whole files, it queries a local index and gets back a handful of tree-sitter chunks. On token accounting alone that math usually works out in your favor. Pick it when the code stays on your machine — regulated environments, client work under NDA, air-gapped setups. The static embedding model means no transformer forward pass at query time, no GPU, no API keys, nothing phoning home. Setup is one command and the installer wires up whichever agents you already have. The reranking is the part worth caring about. Definition boosts push the chunk that actually declares get_user_config above the twelve that merely call it. Stem matching catches the camelCase/snake_case split that trips up plain lexical search. Noise penalties bury test fixtures and .d.ts stubs by default, which is right far more often than not. Where it bites: those same noise penalties are opinionated. If you're debugging a test helper or chasing a type stub, the canonical implementation will outrank what you asked for, and you'll need to steer the query. Retrieval quality is also corpus-dependent — the benchmark NDCG is the vendor's, measured on their repos. Run a few of your own questions before you trust it across a large unfamiliar codebase. Compare it to hosted code search services and you're trading dashboards, access controls, and team analytics for privacy and zero cost. Compare it to an agent's built-in grep tool and you're trading nothing but ten minutes of setup. For a solo dev or a small team already running Cursor or Claude Code locally, that trade is easy. For an org that needs audit logging on who searched what,
Researching Semble? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Semble actually fits — and what changes day-one when you adopt it.
Install uv and Semble, run 'semble install', let it detect Claude Code, and enable the MCP server. Then ask the agent 'How is authentication handled?' instead of letting it grep and read whole files.
Outcome: The agent gets back only the relevant chunks — the vendor reports ~99% fewer tokens than grep+read and ~1 ms query latency — so more of the context window goes to reasoning and less to file dumps.
Register Semble's MCP server and call the search tool with repo set to a local path or an https:// git URL and content set to code, docs, config, or all. Use find_related with a file_path and line number to pull semantically similar code.
Outcome: Code retrieval becomes a two-tool MCP primitive rather than custom glue, and passing a list of repos lets you search several related repositories together.
Run Semble entirely on CPU with no API keys, GPU, or external services. Indexes cache to disk on first search and reindex incrementally as files change; in MCP mode the index refreshes automatically as you edit.
Outcome: Code search stays local while results stay current across the session, with no cloud calls and no dependency on an external embedding endpoint.
Use Cases
- Ask a large codebase a natural-language question like 'How is authentication handled?' and get back only the relevant snippets
- Find semantically related code for a given file and line number with the find_related MCP tool
- Wire code retrieval into Claude Code or Cursor through the MCP server
- Index a git repository by URL and query it without cloning manually
- Search several related repos together in one query
- Restrict a search to docs or config rather than code with the content parameter
- Cut token usage in agent workflows by returning chunks instead of whole files
- Point Semble at a custom Model2Vec-compatible model when Hugging Face isn't reachable at runtime
Models Under the Hood
as of 2026-10-16
Limitations
- Semble runs entirely locally on CPU with no API keys, GPU, or external services; there is no hosted or managed offering implied by the evidence, and no web dashboard, RBAC, audit logging, or team analytics.
- Accuracy figures (NDCG@10 of 0.854) and speed claims (~500 ms indexing, ~1 ms queries, ~380x faster indexing and ~17x faster querying than a code-specialized transformer) are the vendor's own benchmarks, not independent evaluations.
- Getting it into an agent requires installing it and choosing an integration (MCP server, AGENTS.md CLI instructions, or a dedicated sub-agent), which is a developer setup task.
- Noise penalties down-rank test files, compat/ and legacy/ shims, examples, and .d.ts stubs by default — if you need to search those first, that default works against you.
- It is a code search library for agents rather than a full application platform.
as of 2026-10-11
Verification history
We have re-verified Semble 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Semble tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Open Source
$0
Ideal for
Solo devs and small teams running AI coding agents against large private repos who want code search with no cloud calls, no API keys, and no per-seat cost.
What this tier adds
Starting tier and the only published tier: $0, MIT-licensed, with the full retrieval stack, MCP server, CLI and sub-agent integrations, and custom model support.
Where the pricing makes sense
The company stage and team size where Semble's pricing actually pencils out — and where peers do it cheaper.
Semble costs nothing: it's MIT-licensed open source, and the published tier is $0. That makes price comparisons against hosted code-search services straightforward — those bill per seat or per query, while Semble's cost is the engineering time to install it and the CPU you already have. It fits solo devs and small teams running agents on large private repos; it is not aimed at buyers who want to purchase a managed service with SLAs, dashboards, and support contracts.
Setup time & first value
How long it actually takes to get something useful out of Semble — broken out by persona, not the marketing-page minute.
For a developer: minutes, not hours. Install uv, run 'uv tool install semble' and then 'semble install' — the installer detects your coding agents (Claude Code, Cursor, Codex, Gemini CLI, OpenCode, and more) and lets you choose MCP server, AGENTS.md CLI instructions, or a dedicated sub-agent. First search builds the index (vendor benchmark: ~500 ms for an average repo) and caches it to disk.
Switching to or from Semble
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From grep + read-full-file agent loops: install Semble and enable the MCP server so your agent gets chunks instead of whole files
- →From a hosted code-search service: point Semble at a local path or an https:// git URL and keep queries on CPU with no cloud calls
- →From per-repo manual cloning: pass the https:// git URL directly to the search tool instead of cloning first
- →From a single-repo search setup: pass a list of repos to search related repositories together in one query
- ↗To a hosted code-search platform: if you need a web dashboard, role-based access control, audit logging, or team analytics, a managed service is the replacement, not Semble
- ↗To a code-specialized transformer retriever: the vendor benchmarks Semble as matching that quality while indexing ~380x faster and querying ~17x faster, so measure before switching
- ↗To another MCP retrieval tool: run 'semble uninstall' to undo the agent integrations before switching
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Semble”, and we withheld 6: 6 could not be judged, because “Semble” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Semble.
Official links
Tools that pair well with Semble
Common stack mates teams adopt alongside Semble, with the specific reason each pairing earns its keep.
Chrome DevTools MCP
Apache-2.0 MCP server that gives coding agents a live Chrome DevTools session for debugging, traces, and browser automation.
Codeium
Devin Desktop is an agent-native IDE for dispatching and reviewing fleets of local and cloud coding agents.
Windsurf
Devin Desktop (formerly Windsurf): run fleets of local and cloud coding agents from one Sessions board.
Featured Head-to-Head Comparisons
Semble vs Spider Cloud
These tools aren't direct competitors: Spider Cloud excels at web data extraction for AI/LLM pipelines, while Semble is a local code search library for AI coding agents. Choose based on your data source—web or code.
Semble vs Temporal Ai
Choose Temporal if you need robust, durable orchestration for AI agents or microservices that survive failures and require human-in-the-loop; choose Semble if you mainly want to supercharge your AI coding agent with instant, token-efficient code search. They solve different problems — Temporal is a workflow engine, Semble is a code retrieval library — so the right choice depends on whether your bottleneck is reliability or context window limits.
Semble vs Voyage Ai
If you need high-accuracy retrieval on enterprise documents with domain-specific models and compliance, Voyage AI is the clear choice—but be prepared for enterprise pricing. For developers building AI coding agents that need instant, local, and token-cheap code search, Semble is a fantastic free tool that integrates seamlessly with popular IDEs and MCP workflows. Choose based on your primary use case: documents vs. code.
Alternatives to Semble
View all
Chrome DevTools MCP
Apache-2.0 MCP server that gives coding agents a live Chrome DevTools session for debugging, traces, and browser automation.
Frequently Asked Questions
Best-of guides
Used Semble? Help shape our editorial sentiment research.