Vectr

Semantic search and persistent memory for AI code editors.

Version 1.5.0 · Last updated 2026-07-24 · CHANGELOG

Research: the memory design is grounded in our paper Delivery, Not Storage: Cue-Anchored Working Memory as a Harness Property for Coding Agents (arXiv:2607.20972).

Vectr gives AI code editors two things they lack: semantic codebase search and persistent working memory — both served over MCP with zero configuration.

Your AI editor forgets everything. Vectr doesn't.

The problem

Every time an AI code editor starts a task, it re-reads the same files it read yesterday. On an unfamiliar codebase it runs ripgrep, reads entire files hunting for the right function, and fills its context window with noise. In a long session it loses findings from turn 1 by turn 40. Across sessions it starts over from zero.

Vectr breaks the re-discovery loop:

• One index → semantic search over your whole codebase in <20ms
• One recall call → structured notes from any prior session, verbatim, in <50ms
• Survives /compact → notes are persisted to disk, not stored in context

Measured, not hypothetical: recalling 3 stored notes with vectr_recall costs 360 tokens in one tool call. Re-deriving the same three facts with grep + Read costs ~2,060 tokens across six tool calls on the same 182-file Python repo — ~5.7× fewer tokens, 6× fewer tool calls, in under 50ms (chars/4 tokenization; full breakdown in Measured costs, honestly). Across a 6-task CPython sprint measuring real Read+Bash calls, that recall discipline cut re-discovery by 39% overall, with per-task reductions ranging 0%–85% depending on how unfamiliar the code was to the model (the 0% task is one the model could already navigate from training — see When vectr can hurt).

Notes are persisted to disk, not held in the conversation — they survive /compact and a fresh session equally; the session boundary doesn't matter.

No API key required. The embedding model runs locally.

Benchmarks — CPython internals sprint (6 tasks, 2 agents)

The benchmark simulates a week of feature work on an unfamiliar C codebase (CPython internals). One research session stores findings with vectr_remember; six isolated implementation sessions each open cold and call vectr_recall.

Implementation sessions only — 6 tasks combined:

Per-task re-discovery (Read+Bash before first write):

Research vs implementation cost breakdown:

The research phase (paid once to build notes) costs more for vectr (+94%) because it stores rich code stubs and function signatures via vectr_remember. The implementation phases (which repeat every task) cost less because vectr_recall replaces file re-discovery. The research overhead breaks even after ~8 tasks of note reuse.

Earlier runs on Apache Camel (Java, 5,856 files): −58% impl cost · −72% impl tool calls · −39% wall time.

Full results: benchmarks/

Measured costs, honestly

Per-call token cost (median, 182-file Python repo, chars/4 tokenization):

The trade-off, stated plainly: for a single pointed lookup on a small, already-familiar repo, grep is cheaper — vectr's median cost across 5 single-fact tasks was +60% more tokens — and faster, since a vectr_search round-trip takes 1.7–3.6s against ~28ms for grep. Vectr doesn't win on per-call cost; it wins on tool-call count (one round-trip instead of several), answer completeness (a whole symbol back, not a partial file read), and everything in working memory — the 5.7× recall refund from the opening section compounds with every task you resume.

Fine print: the automatic eviction/reminder banners riding along on tool responses cost tokens too — an always-on re-fetch footer runs ~27 tokens, a light nudge ~89 tokens, and the escalated action-required banner (fires only after both the chunk and token thresholds are crossed without a save) scales from ~480 to ~535 tokens before it plateaus.

When it pays off: unfamiliar or large codebases, work you resume (later this session, after /compact, or in a new session), and long sessions with many turns. When it doesn't: a one-off grep on code you already know cold — reach for grep instead.

Quick start

Local (recommended)

python3.14 -m venv ~/.vectr-env
source ~/.vectr-env/bin/activate   # Windows: ~/.vectr-env/Scripts/activate
pip install vectr
cd /path/to/your/project
vectr start

Requires Python 3.14+. To install:

• macOS: brew install python@3.14
• Ubuntu/Debian: sudo add-apt-repository ppa:deadsnakes/ppa && sudo apt install python3.14 python3.14-venv
• Windows: python.org/downloads

vectr start returns immediately. Indexing runs in the background — run vectr status to check progress. On first run the embedding model downloads once (~290 MB). Restart your AI code editor once to pick up the new MCP config.

Docker (CI/servers)

git clone https://github.com/swapnanil/vectr
cd vectr
docker-compose up api

Exposes port 8765. Docker does not auto-write IDE config files — use local install for IDE integration.

Connect to your AI code editor

vectr start writes the MCP config for your editor automatically. Restart your editor once.

"Verified" means the full integration (config, guidance, and hooks) has been exercised end to end. "Experimental" means the MCP config is written and works, but the integration hasn't been run through the same verification pass. "Planned" means no support yet.

{ "mcpServers": { "vectr": { "type": "http", "url": "http://localhost:8765/mcp" } } }

{ "mcpServers": { "vectr": { "url": "http://localhost:8765/mcp" } } }

{ "servers": { "vectr": { "type": "http", "url": "http://localhost:8765/mcp" } } }

{ "mcpServers": { "vectr": { "serverUrl": "http://localhost:8765/mcp" } } }

{ "mcpServers": [{ "name": "vectr", "transport": { "type": "http", "url": "http://localhost:8765/mcp" } }] }

[mcp_servers.vectr]
url = "http://localhost:8765/mcp"

Stdio transport

The editors above connect over HTTP (vectr start + POST /mcp). For MCP clients and hosting platforms that spawn the server as a subprocess and speak MCP over its stdin/stdout instead of opening an HTTP connection, run:

vectr mcp-stdio [WORKSPACE]

No port, no daemon — a single foreground process framed as newline-delimited JSON-RPC 2.0 (one JSON object per line on each of stdin/stdout, no Content-Length headers). initialize and tools/list answer immediately; the embedding model load and initial indexing happen on a background thread, and a tools/call made before that finishes gets a graceful "still starting up" response instead of hanging. Stdout carries protocol JSON only — all logging goes to stderr. The process exits cleanly on stdin EOF (the client closing the pipe). --memory-only and --search-only behave the same as on vectr start.

How it works

1. AST-aware chunking — tree-sitter parses each file and splits at function/class/method boundaries. No chunk breaks mid-logic.
2. Code embeddings — ibm-granite/granite-embedding-english-r2 (local, CPU-fast, overridable) maps natural-language queries to code symbols ("JWT validation" → verify_jwt_token). BM25 handles exact symbol names.
3. Hybrid search — vector similarity + BM25 combined, weighted by codebase characteristics (large/unfamiliar → semantic-heavy; small/well-documented → BM25-heavy).
4. Symbol graph — call edges, import chains, and HTTP routes (Flask/FastAPI/Express/Spring) are extracted and stored. vectr_locate uses 5 fallback strategies: exact match → suffix → same-module → unique-name → import-chain → fuzzy (edit distance ≤ 2).
5. Working memory — vectr_remember stores structured notes to SQLite + ChromaDB. vectr_recall does semantic search over notes — not SQL LIKE — so multi-word queries always find relevant context.
6. MCP protocol — 19 tools served over HTTP. Any MCP-compatible AI code editor connects without plugins.

19 MCP tools

vectr start writes a CLAUDE.md into your workspace with this table and usage guidance — your AI code editor knows which tool to reach for without being prompted.

Search tools — retrieve code from the index:

Memory tools — store and recall across sessions:

Workspace-scoped notes double as a shared bus for multi-agent workflows: an orchestrator and its subagents all read and write the same note store, so a subagent can call vectr_remember(..., agent="coder-2") with its findings before finishing, and the orchestrator recalls them instead of re-reading the subagent's full transcript. The agent param is never inferred — it's explicit attribution, and it shows up as a tag in vectr_recall index output.

On editors with session hooks (see the host-support matrix for which ones), recall is injected automatically — directives and high-priority tasks at session start, semantic recall keyed to each prompt, and file-anchored gotchas before a read or edit — with observability via a Hook injections line in vectr status.

Those are the kind defaults. A note can also carry explicit per-note triggers on vectr_remember: path globs, lifecycle events (session-start, prompt-submit, pre-edit, pre-run, pre-commit, post-compaction), exact symbol references resolved against the same symbol graph vectr_locate uses, a semantic prompt-similarity match with a fixed per-kind threshold, and temporal guards (not_before, expires_visibility, cooldown). Conditions AND within an entry and OR across entries, evaluated deterministically by the daemon with a per-session fire ledger and injection budgets — so a note resurfaces exactly when its condition holds, and never twice in the same window.

CLI reference

vectr start                           # index + start daemon for current dir
vectr start /project/api              # positional workspace: a directory or .code-workspace file
vectr start --path /project/api       # specific workspace (repeatable, multi-root)
vectr start --memory-only             # working memory + hooks only — no code index, no watcher
vectr status                          # index health, chunk count, notes count
vectr status --all                    # all running instances
vectr stop /project/api               # stop one instance (same positional as start)
vectr stop --path /project/api        # stop one instance (equivalent --path form)
vectr stop --all                      # stop all instances
vectr index --path .                  # re-index without restarting daemon
vectr fetch src/auth.py:10-42         # re-fetch a chunk by exact id, verbatim
vectr init --path .                   # write CLAUDE.md + MCP config without starting
vectr init --exclude vendor           # exclude directories from indexing
vectr forget --path .                 # delete all working-memory notes
vectr cache prune                     # remove empty per-workspace cache dirs (live instances skipped)
vectr cache prune --dry-run           # preview what would be removed, delete nothing
vectr proxy                           # experimental: localhost API proxy (see below)
vectr mcp-stdio                       # foreground stdio MCP transport, no port/daemon (see above)

Excluding paths

Create .vectrignore in your project root (same syntax as .gitignore):

vendor/
node_modules/
*.pb.go
dist/

Or pass --exclude at init time:

vectr init --exclude vendor --exclude dist

Exclusions apply to both the initial index walk and the live file watcher, so adding a directory to .vectrignore stops a running instance from re-indexing it. The next index also prunes any chunks already stored for now-excluded (or deleted) files — you don't have to rebuild from scratch. If you ever need a clean rebuild (e.g. after changing the embedding model), force one:

vectr index --path . --force      # ignore the incremental cache, re-embed everything

Supported languages

HTTP routes (Flask/FastAPI decorators, Express app.get(), Spring @GetMapping) are extracted as symbols and searchable via vectr_locate("GET /api/users").

Cost

Security

The default is unchanged and stays the headline: local, no API key, zero config — a solo developer on a personal machine. Out of the box, the daemon binds to 127.0.0.1 only, CORS is restricted to localhost origins, each workspace gets its own isolated DB directory, port, and process (owner-only 0700 on POSIX), and the index and notes persist locally in ~/.cache/vectr/.

Everything below is opt-in; enabling nothing changes nothing.

Authentication — set VECTR_API_KEY and every request to /v1/* and /mcp must carry it (X-Api-Key: <key> or Authorization: Bearer <key>; constant-time comparison; /v1/health stays open for liveness probes). Generate a key with vectr key. When the key is set at start time, the editor MCP configs vectr writes include the header automatically. Those files (.mcp.json, .cursor/mcp.json, .vscode/mcp.json) then hold the key in plaintext — treat them as secrets and keep them out of shared or public version control.

Encryption at rest — set VECTR_ENCRYPT_KEY (or store a passphrase in the OS keychain: service vectr, username encrypt-key; requires pip install vectr[encryption]) and note content, note titles, and snapshot payloads are encrypted (Fernet, PBKDF2-derived key). Honest boundary: the code index is not encrypted — the search engine needs readable chunk text and vectors; protect it with OS full-disk encryption. Note tags/metadata stay plaintext, and note embedding vectors (a lossy projection of note text) are kept for semantic recall unless you set VECTR_ENCRYPT_DISABLE_NOTE_VECTORS=1.

Retention and audit — notes are kept until you delete them; set VECTR_NOTES_TTL_DAYS to auto-purge older notes at startup. vectr_forget(all=true) / vectr forget --all delete notes, snapshots, and note vectors — everything means everything. Set VECTR_AUDIT_LOG=<path> for a rotating local log of index/search/remember/recall events (off by default; it records query text — that is its purpose — and is never transmitted). Full policy: docs/data-handling.md.

Team mode (shared instance) — one central daemon can serve a team on one repo: VECTR_API_KEY=<key> vectr start --host 0.0.0.0 on the server (a non-loopback bind refuses to start without a key), then vectr connect --url http://<host>:<port> --api-key <key> --label <you> on each client to point the editor at it. Working memory is shared: a note one agent stores, every connected agent can recall; --label attributes notes and audit lines. Note IDs are allocated by the central store in the order writes arrive, so with concurrent clients your notes are not a contiguous block — they interleave with other clients'. The Stored note #N line the write returns is the canonical reference to that specific note; don't assume the next ID is also yours. Plain limits: one shared key means every holder is an equal, trusted peer (no roles, no per-user permissions); the server operator can read everything; search results reference the server's checkout, which may differ from your local tree; vectr speaks plain HTTP — put TLS at a reverse proxy or tunnel if the network isn't trusted.

Proactive context (experimental, off by default)

Pull-based recall means the agent has to ask. Proactive context is the opposite: the right note arrives exactly when it is relevant. It is experimental, off by default, and localhost-only.

The vectr proxy command runs a small local proxy that sits on the wire between your agent and the model API:

vectr proxy                            # starts a localhost proxy (default :8785)
export ANTHROPIC_BASE_URL=http://127.0.0.1:8785   # point your agent at it

What it does and does not do, plainly:

• Transparent by default. It forwards every request to the real Anthropic API — streaming responses and tool calls pass through byte-for-byte. Your API key is forwarded untouched and never stored or logged.
• Deterministic injection, when enabled. With VECTR_PROACTIVE=1 and the workspace daemon running, it appends matched working-memory notes to the request after the last prompt-cache breakpoint, so your prompt cache is never invalidated. Triggering is a similarity threshold + exact structural matches — never keyword guessing — with a strict per-request budget so a hint only lands when it is worth the tokens.
• Fail-open. If the intelligence layer is slow or errors, your request goes through unchanged. To bypass the proxy entirely, unset the base URL: unset ANTHROPIC_BASE_URL. The proxy is on the request path, so if it is down, unset the variable to talk to the API directly.
• Solo/localhost-only. It reads your conversation to compute context, so it refuses any non-loopback bind and is mutually exclusive with team mode.
• Caveats on a non-first-party base URL (from the Claude Code docs): MCP tool search is disabled unless ENABLE_TOOL_SEARCH=true and the proxy forwards tool-reference blocks; Remote Control is disabled on a non-api.anthropic.com base URL.

Org-wide caching (team mode). With a central shared instance, vectr can cache its own expensive artifacts — semantic search and recall results — keyed by exact identity and the current index state, so a re-index or note change invalidates them automatically and every connected developer benefits. It is off by default (proactive.cache), and vectr status reports its hit rate and estimated tokens saved so the value is measured, not asserted. vectr does not cache LLM responses across similar requests — only byte-identical ones, locally, opt-in — because a wrong cache hit would silently corrupt a conversation.

When vectr can hurt

Stale notes after codebase churn — notes store file paths at write time. After a large refactor, vectr_recall will flag changed referenced files with [STALE]. Re-verify before acting, delete the stale note with vectr_forget(note_id=N), or clear everything with vectr_forget(all=true).

Over-retrieval on a well-known API — if the model already knows a framework deeply from training (React hooks, Django ORM), vectr's research overhead may exceed savings. The benchmark shows 0% improvement on debug_descriptor_priority — a task where the model's training knowledge was sufficient to navigate without notes.

Built with

Python 3.14 · FastAPI · sentence-transformers · tree-sitter · ChromaDB · BM25 · Docker

Author

Swapnanil Saha · swapnanilsaha.com