archex 0.13.0

Architecture extraction & codebase intelligence for the agentic era

archex

---

Verified local code context for agents.

archex turns a repository into a ranked, token-budgeted context bundle plus a context receipt with freshness, index revision, skipped candidates, omitted dependency edges, and a recommended next action. It runs locally, uses deterministic retrieval and analysis, and does not require hosted inference or an API key. The v0.13 line adds stronger benchmark trust surfaces, bundle-only evaluator support, and default 4-bit TurboQuant vector storage for local vector indexes.

Start: 30-second quickstart · MCP and Claude Code · Python API · Local metrics · Compatibility matrix · Installation trust contract · Security policy

Quick links: Proof bar · Fast paths · What archex returns · Use it your way · Trust and operations · Measured results · Advanced workflows · Installation details · Language support · Development · Documentation map

Watch the explainer · Open banner SVG · Open infographic SVG · Read the measured comparison

Proof bar

Safe-to-act signals	Surfaces	Language coverage	Public evidence
Query/scout receipts expose freshness, index revision, skipped candidates, omitted edges, completeness, and next action	CLI, MCP, Python API, Docker, Claude Code skill	25 declared language IDs with explicit full vs chunk-only tiers	C1 public comparison, raw-ripgrep/read baseline, bundle-only evaluator lane, and TurboQuant A/B measurement with 7.07× mean vector .npz compression

archex does not ask the downstream agent to trust ranking alone. Every query/scout receipt explains what was returned, what was skipped, whether freshness was current, and whether the bundle is complete enough to act on.

Fast paths

If you are evaluating...	Start here	Why
Agent workflows	archex doctor, then archex scout "question" --budget 1000 --format json	Checks local trust first, then returns a compact map, a receipt summary, and exact fetch handles.
Claude Code or MCP	MCP and Claude Code	Stdio MCP server, optional warm --watch, additive top-level receipts, and an in-repo skill that teaches doctor → scout → fetch.
Python applications	Python API	Deterministic query(), analyze(), compare(), and receipt-bearing bundles.
Benchmark proof	Measured results and archex vs. cocoindex-code	Same-task C1 report, raw-ripgrep/read baseline, bundle-only evaluator reports, required-file trust gates, and TurboQuant storage/recall evidence.
Installation and clients	Compatibility matrix	Preview-first client bootstrap paths for Claude Code, Codex, Pi, OpenCode, and Cursor.

30-second quickstart

uv tool install archex
archex doctor
archex query "How does authentication work?" --format xml

archex doctor reports whether the local index, grammar support, model cache, MCP registration, and .archex/ state are healthy. Repo-local commands default to the current working directory. If the repo has not been initialized yet:

archex init
archex index
archex query "How does authentication work?" --format xml

What archex returns

archex returns a context bundle plus receipt, not an answer. The downstream agent or model still does the reasoning; archex decides which code, symbols, dependencies, and type context belong in the prompt, then records why that bundle is safe or incomplete.

<context query="How does authentication work?">
  <structural-context>
    <file-tree><![CDATA[
src/auth/
  middleware.py
  tokens.py
  models.py
    ]]></file-tree>
  </structural-context>
  <chunks>
    <chunk file="src/auth/middleware.py" lines="42-78" symbol="authenticate" score="0.9312" tokens="284">
      <imports><![CDATA[from auth.tokens import verify_jwt]]></imports>
      <code><![CDATA[
def authenticate(request: Request) -> User:
    token = extract_bearer(request)
    claims = verify_jwt(token)
    return load_user(claims.sub)
      ]]></code>
    </chunk>
  </chunks>
  <type-definitions>
    <type-def file="src/auth/models.py" symbol="User" lines="10-24"><![CDATA[
@dataclass
class User: ...
    ]]></type-def>
  </type-definitions>
  <dependencies>
    <internal>auth.tokens.verify_jwt</internal>
    <external>pyjwt</external>
  </dependencies>
</context>

The bundle carries ranked chunks, import context, referenced type definitions, dependency edges, token counts, and provenance. Use --format json or --format markdown when XML is not the right downstream envelope.

Small receipt example:

{
  "receipt": {
    "freshness": "clean",
    "index_revision": "3d8b0c…",
    "token_budget": { "requested": 12000, "consumed": 6132 },
    "returned_total": 12,
    "skipped_total": 23,
    "included_edges_total": 9,
    "omitted_edges_total": 17,
    "context_complete": "incomplete",
    "context_complete_reason": "dependency_frontier_cut",
    "recommended_next_action": "fetch_skipped_candidate",
    "returned_context": [
      {
        "handle": "chunk:src/auth/middleware.py::authenticate#function",
        "file_path": "src/auth/middleware.py",
        "start_line": 42,
        "end_line": 78,
        "score": 0.9312
      }
    ],
    "skipped_candidates": [
      { "file_path": "src/auth/session.py", "reason": "below_threshold" }
    ]
  }
}

Use CONTEXT_RECEIPTS for the full field contract.

Why archex is different

Agents usually explore repositories by opening one file, following imports, checking type definitions, and backtracking. That burns context before the real task starts. archex performs local retrieval and structural expansion first: BM25F, optional local vector/SPLADE signals, graph expansion with edge confidence, type-definition packing, and intent-routed token budgets.

Repository → repo-local index → intent routing → retrieval → graph/type expansion → token-budgeted bundle → agent / MCP client

archex is a selection and assembly layer. Compression tools can shrink the final bundle later, but compressed irrelevant context is still irrelevant. For the vector index itself, v0.13 enables 4-bit TurboQuant storage by default when vector retrieval is turned on: same measured recall/MRR on the current corpus, about seven times smaller vector artifacts, and self-describing compatibility with older unquantized .npz files.

Use it your way

CLI

archex query "Where is cache invalidation handled?" --format xml
archex scout "How does authentication flow through this repo?" --budget 1000 --format json
archex index --quantize-vectors --quantize-bits 4 --allow-remote-code
archex graph export --output .archex/archgraph.json
archex graph neighbors src/auth/middleware.py --graph .archex/archgraph.json --format markdown
archex symbol 'symbol:src/auth/middleware.py::authenticate#function'

MCP and Claude Code

Install the MCP extra and register the stdio server:

uv tool install "archex[mcp]"

{
  "mcpServers": {
    "archex": { "command": "archex", "args": ["mcp"] }
  }
}

Preview the exact client config before writing it:

archex install-client claude-code .
archex install-client claude-code . --write

For warm local sessions, keep the MCP process alive and optionally watch the repo:

archex mcp --watch --watch-path .

The in-repo Claude Code skill lives at skills/archex/. Its /archex command runs archex doctor, initializes/indexes when needed, scouts first for broad questions, then fetches exact symbol: or chunk: handles before a larger bundle query.

Exact install, MCP, Docker, cache, uninstall, and trust semantics are documented in the installation trust contract. Client-specific config targets and preview-first bootstrap paths live in the compatibility matrix.

Local usage metrics are off by default. If a user explicitly enables them with archex metrics enable, ARCHEX_USAGE_METRICS=on, or the persisted metrics setting, archex writes a machine-local ledger at ~/.archex/usage.sqlite. That ledger records anonymous counters only: tool name, category, token counts, file count, repo-local random ID, freshness, and index revision. It does not store query text, file paths, symbols, handles, rendered outputs, prompt bodies, remote URLs, org names, or repo names in event rows. Headline savings are always tokens_saved = max(returned full-file baseline - returned tokens, 0). Whole-repo avoided tokens are tracked separately as an upper-bound/context metric when the indexed repo total is available.

Important boundary: archex ships with no telemetry by default. Optional local metrics are separate from telemetry, stay on the machine, and require explicit enablement. Detailed traces remain a second explicit opt-in on top of metrics enablement. The exact calculation rules, privacy boundary, and controls live in LOCAL_METRICS.

archex metrics is the control surface:

archex metrics enable
archex metrics
archex metrics export --output usage.json
archex metrics delete --all
archex metrics trace enable
ARCHEX_USAGE_METRICS=on archex query "Where is auth handled?"

Detailed traces stay opt-in via archex metrics trace enable or ARCHEX_USAGE_TRACE=on. Traces remain local-only and still never store source code or rendered outputs. Metrics code paths make no LLM calls, no hosted upload calls, and no background network calls in v1.

Python API

from archex import query
from archex.models import RepoSource

bundle = query(
    RepoSource(local_path="."),
    "Where is database connection pooling implemented?",
)
print(bundle.to_prompt(format="xml"))

analyze() returns an ArchProfile; compare() returns deterministic cross-repo dimension comparisons. LangChain and LlamaIndex retrievers ship as optional extras.

Docker

Two local-first images are built in CI:

# BM25-only, no torch
docker run --rm -v "$PWD:/workspace" -w /workspace ghcr.io/mathews-tom/archex:slim archex doctor

# Full local-embedding image with FastEmbed runtime
docker run --rm -v "$PWD:/workspace" -w /workspace ghcr.io/mathews-tom/archex:full archex query "Where is cache invalidation handled?" --strategy hybrid

Warm-container MCP pattern:

docker run -d --name archex-mcp -v "$PWD:/workspace" -w /workspace ghcr.io/mathews-tom/archex:slim sleep infinity
docker exec -i archex-mcp archex mcp

MCP client config for that container:

{
  "mcpServers": {
    "archex": {
      "command": "docker",
      "args": ["exec", "-i", "archex-mcp", "archex", "mcp"]
    }
  }
}

The mounted repository owns .archex/, so indexes survive container restarts and stay out of source control.

Trust and operations

Surface	Contract
Security policy	Supported versions, disclosure workflow, no-telemetry posture, secret-handling guidance, and model remote-code policy live in SECURITY.
Context receipts	Field contract, freshness/completeness semantics, output surfaces, and benchmark linkage live in CONTEXT_RECEIPTS.
Compatibility matrix	Tested vs unverified clients, exact config shapes, preview-first bootstrap commands, and verification steps live in CLIENT_COMPATIBILITY_MATRIX.
Installation trust contract	Exact CLI, MCP, Docker, skill, cache, network, freshness, benchmark, and uninstall semantics live in INSTALLATION_TRUST_CONTRACT.
archex install-client	Preview-first client config writer for Claude Code, Codex, Pi, OpenCode, and Cursor.
archex doctor	Text/JSON diagnostics for index health, staleness, local model cache presence, grammar availability by tier, MCP registration, model security, and .archex/ disk usage.
Repo-local .archex/	Generated state: settings, metadata, SQLite index, optional vectors, graph artifacts, dogfood history. Keep it uncommitted.
Local usage metrics	Calculation rules, privacy boundaries, default-off versus opt-in behavior, export/delete controls, and retention live in LOCAL_METRICS.

Measured results

The public C1 harness publishes the same external-repo comparison for archex, cocoindex-code (ccc), and a raw-ripgrep/read baseline. It records cold-start, warm latency, recall, precision, F1, token efficiency, required-file recall, missed-required-file rate, missed-required-task rate, all-required-present rate, receipt accuracy, and bundle-completion penalty tokens. The checked-in artifacts include those trust fields; receipt accuracy is n/a for the historical C1 run because those artifacts predate query receipt capture. Core retrieval benchmarks make no LLM calls.

See archex vs. cocoindex-code for the current published comparison and Retrieval Default Decisions for the decision trail.

Bundle-only evaluation is a separate opt-in lane: archex benchmark bundle-eval --evaluator-command ... gives a user-supplied local command only the rendered bundle and receipt JSON, then reports bundle-only success and files the evaluator still needed outside returned context. archex does not provide hosted evaluator calls, telemetry, credentials, or default network behavior for that lane.

TurboQuant evidence is measured separately with archex_query_hybrid_quantized_4bit against archex_query_hybrid: 35 tasks, 7.07× mean vector .npz compression, 6.98× minimum compression, recall Δ +0.000, MRR Δ +0.000, F1 Δ +0.000, required-file recall Δ +0.000, and mean query latency Δ +110 ms. That passed the default gate, so 4-bit TurboQuant is now the default storage mode for vector indexes.

Lane	Recall	Required-file recall	Missed task rate	F1	Token efficiency	Token efficiency after completion	Warm latency ms
archex	0.95	0.95	0.16	0.66	0.76	0.74	408
ccc	0.32	0.32	0.79	0.31	0.48	0.41	521
raw-ripgrep/read	1.00	1.00	0.00	0.05	0.00	0.00	773

Advanced workflows

# Repo-local lifecycle
archex init
archex index
archex status --strict
archex doctor --format json

# Architecture and graph surfaces
archex analyze --format markdown
archex onboard
archex graph export --output .archex/archgraph.json
archex graph path src/archex/cli/query_cmd.py src/archex/serve/context.py --graph .archex/archgraph.json --format markdown
archex impact --changed-file src/archex/serve/context.py

# Benchmarks and gates
archex benchmark headtohead report --input .archex/headtohead --format markdown
archex benchmark run --strategy archex_query_hybrid_quantized_4bit --output .archex/e2e-quantized --allow-remote-code
archex benchmark report --input .archex/e2e-quantized --baseline .archex/e2e-baseline --format markdown
archex benchmark gate --input .archex/e2e --baseline .archex/e2e-baseline --warn-latency-ms 3000
archex benchmark bundle-eval --tasks-dir benchmarks/tasks --evaluator-command ./local-evaluator
archex dogfood --all --baseline benchmarks/dogfood_baseline.json --format dogfood-delta

Installation details

uv tool install archex                    # CLI, system-wide
uv add archex                             # project dependency

# Agent integrations
uv tool install "archex[mcp]"             # MCP server
uv add "archex[langchain]"                # LangChain retriever
uv add "archex[llamaindex]"               # LlamaIndex retriever
uv add "archex[lsap]"                     # LSP type enrichment

# Local retrieval extras
uv add "archex[vector-fast]"              # FastEmbed (ONNX-backed, ~50MB)
uv add "archex[vector-torch]"             # sentence-transformers / torch
uv add "archex[splade]"                   # SPLADE sparse retrieval
uv add "archex[graph]"                    # Leiden graph clustering
# Core extras bundle: graph, MCP, LangChain, LlamaIndex
uv add "archex[all]"

For the full trust contract, including exact MCP JSON, Docker commands, cache locations, network behavior, and uninstall steps, see Installation and Trust Contract.

Language support

Tier	Languages	Extraction
full	Python, JavaScript, TypeScript/TSX, Go, Rust, Java, Kotlin, C#, Swift	Symbols, imports, graph edges
chunk-only	C, C++, PHP, Ruby, Scala, Lua, Bash/Shell, SQL, HTML, CSS, YAML, TOML, JSON, Markdown, Solidity	AST chunking + retrieval; no symbol/import graph claim
unknown	any other text file	line-window chunks for BM25 visibility

Need another language? Register an adapter via Python entry points. See System Design for the extension contract.

What archex is not

• Not a chatbot — it emits context bundles; another agent or LLM does the explaining.
• Not a hosted RAG service — indexing and retrieval run locally unless you explicitly query a remote Git URL.
• Not a vector database — vector search is optional; BM25 and structural signals are first-class.
• Not an LSP replacement — use LSAP/LSP where compiler-backed type resolution matters; archex packages repository-scale context for agents.
• Not a prompt template library — output is structured retrieval evidence, not prompt prose.

Development

git clone https://github.com/Mathews-Tom/archex.git
cd archex
uv sync --all-extras

uv run ruff check && uv run ruff format --check .
uv run pyright
uv run pytest

Documentation map

Authority chain: README → System Design / archex vs. cocoindex-code → Roadmap completion record → Retrieval Default Decisions.

• Why archex — the agent token problem this solves
• System Overview — current product overview and boundaries
• System Design — shipped architecture, graph query, scout, language tiers, and distribution surfaces
• archex vs. cocoindex-code — evidence-backed C1 comparison
• Retrieval Default Decisions — default-strategy and TurboQuant evidence gates
• Context Receipts — receipt field contract and safe-to-act semantics
• Local Metrics — token-savings math, privacy boundary, and default-off versus opt-in behavior

License

Apache 2.0 — see LICENSE.

Star History