Tessera Code Graph

A local, deterministic semantic code graph for AI coding agents.
Stop burning tokens on grep + file reads. Answer “where is this?”, “who calls it?”, “how does A reach B?”, and “did the model hallucinate this symbol?” — deterministically, over a Tree-sitter graph indexed in seconds and queried in milliseconds.

⭐ Star this repo if it saves your agent from a grep spiral — it helps others find it. 💚 Sponsor on GitHub if Tessera saves you tokens at work.

11 languages · deterministic, AST-exact graph · CLI · MCP server · Rust library. Not LLM-extracted — same input, same graph, zero tokens to build it.

---

Measured on real production repos

tessera bench --path . runs against any repo and prints the chart below. The harness ships in the binary — every number here is reproducible.

951-file Java Service

Tessera v0.3.1 bench
─────────────────────
  951 files · 16,368 symbols · 129,959 references

Index time
  full         ████████████████████████████████    2,981 ms
  incremental  █                                      40 ms   ·  75× faster

"who calls parseFrom?"
  raw grep + read   ████████████████████████████████   394,140 tokens
  tessera           █                                    6,530 tokens   ·  60× cheaper

Per-query latency  ·  median of 3 runs
  find_definition      1 ms     ~ 1,781 tokens
  find_references      8 ms     ~16,144 tokens
  impact             371 ms     ~ 6,530 tokens
  validate             1 ms     ~    48 tokens

1,063-file Node.js service

Tessera v0.3.1 bench
─────────────────────
  1,063 files · 3,067 symbols · 142,337 references

Index time
  full         ████████████████████████████████    1,557 ms
  incremental  █                                      38 ms   ·  41× faster

"who calls BaseWorker?"
  raw grep + read   ████████████████████████████████    36,311 tokens
  tessera           █                                       41 tokens   ·  886× cheaper

"where is BaseWorker defined?"
  raw grep + read   ████████████████████████████████     1,790 tokens
  tessera           ██                                      90 tokens   ·  20× cheaper

Per-query latency  ·  median of 3 runs
  find_definition      0 ms     ~    90 tokens
  find_references      8 ms     ~    33 tokens
  impact               1 ms     ~    41 tokens
  validate             0 ms     ~    49 tokens

Headlines

• 60–900× fewer tokens to answer "who calls this?" — the work your agent spends most of its context window on.
• 38–40 ms incremental re-index on near-million-LOC repos — fast enough to run on every file save.
• Sub-20 ms for definition / reference / validation queries.
• CommonJS-aware: require('./foo') is indexed alongside ES6 import, so imports / imported_by work on legacy Node code too.

Install

No Rust toolchain required — pick whatever fits your stack:

npm install -g tessera-codegraph                 # Node / any JS-agent setup (npx tessera-codegraph also works)
brew install iamsaquib8/tessera/tessera          # macOS / Linux (Homebrew)
curl -fsSL https://raw.githubusercontent.com/iamsaquib8/tessera/main/install.sh | sh   # prebuilt binary
cargo install tessera-codegraph                  # from source
docker run --rm -v "$PWD:/work" ghcr.io/iamsaquib8/tessera index /work   # container

Prebuilt binaries for macOS (arm64/x64), Linux (x64/arm64), and Windows are attached to every release.

Zero-install: drop-in agent skill

Prefer not to install anything? Copy the /tessera Agent Skill into ~/.claude/skills/ and Claude Code (or any skill-aware agent) will use Tessera for navigation automatically — installing the binary on first use:

cp -r skills/tessera ~/.claude/skills/tessera

Daily-use commands

tessera init --mcp-configs               # generate project-local defaults and MCP snippets
tessera index .                          # index your repo into .tessera/tessera.db
tessera watch .                          # keep the graph fresh while you edit
tessera doctor                           # check DB, schema, snapshot, parsers, and MCP command
tessera impact findById                  # transitive callers ranked by personalised PageRank
tessera validate findByIdd               # "did the model hallucinate this?" — yes; meant findById (0.98)
tessera connect handleRequest writeRow   # the shortest call path from A to B
tessera export --from findById --format mermaid   # the call subgraph, as a diagram you can paste anywhere
tessera export --from findById --group-by language --html-out graph.html
tessera context-pack findById            # body + deps + callers + tests in one budgeted bundle
tessera plan-query "edit findById safely" --symbol findById
tessera edit-prep findById
tessera unused --kind function           # symbols with no inbound refs/call edges
tessera completions zsh                  # shell completions for bash/zsh/fish/PowerShell
tessera mcp-http --addr 127.0.0.1:8765   # HTTP/SSE MCP transport with /health

That's it. The graph is local, the queries are deterministic, every response carries _meta token estimates so agents can plan their context budget.

Why a deterministic graph?

A wave of tools build a code "knowledge graph" with an LLM extraction pass. That's great for breadth and prose ("what is this and why was it designed this way") — but it's a strange foundation for the one job that matters most to an agent: not getting lied to. An LLM-extracted graph is non-deterministic, costs tokens every time you build it, and can hallucinate edges of its own.

Tessera goes the other way. The graph is pure Tree-sitter AST + static resolution — same input, same graph, every run, zero LLM tokens to build. "Who calls parseFrom?" is a fact from the parser, not an inference. And because the graph is ground truth, Tessera can do the inverse of hallucinating: it catches the model's hallucinations (validate), deterministically. It's also a real engine — CLI, library, and MCP server you can run in CI or a pipeline — not a prompt package that only lives inside an agent session.

graphify-style tools = understand my whole project, explained by an LLM (broad, multimodal, interpretive). Tessera = navigate my code with compiler-grade precision, verified against ground truth, with a hallucination guard (exact, deep, trustworthy).

Different jobs. If you want the second one, this is the tool.

How it compares

	Tessera	LLM-extracted graph (graphify-style)	aider's repomap	Sourcegraph	Cursor's index
How the graph is built	Tree-sitter AST (deterministic)	LLM semantic extraction	static repomap	indexers	proprietary
Tokens to build / maintain the graph	0	LLM pass every build	low	n/a	n/a
Same input → same graph	✅	❌ (non-deterministic)	✅	✅	—
Catches model hallucinations	✅ validate	❌ (can introduce them)	❌	❌	❌
Personalised PageRank impact	✅	❌	✅ (non-personalised)	❌	❌
Call-path tracing (connect) + graph export	✅	❌	❌	partial	❌
Runnable as CLI / library / CI (not just in-agent)	✅	❌ (prompt package)	partial	✅	❌
Local-only, no cloud	✅	✅	✅	❌ (enterprise)	❌
MCP server	✅	varies	❌	❌	❌
Incremental re-index in ms	✅	❌	partial	❌	proprietary
Open source (Apache-2.0)	✅	varies	✅	core	❌

What makes it different

• Personalised PageRank impact. Not just "who calls X" — who calls X that matters. The random surfer teleports back to your edit site, so transitively reachable hubs float to the top with auditable breakdowns (pagerank, fanout_in, fanout_out, exported, test_coverage, depth_decay).

• Hallucination validator. Bloom-filter-fronted symbol existence check + a snippet validator that parses LLM output with the same Tree-sitter pipeline that built the graph. Every call is verified; near-misses come back with Jaro-Winkler confidence scores.

  echo 'findByIdd(1)' | tessera validate-snippet --language typescript
  

  ✗ findByIdd at line 1 col 1
        -> maybe findById (0.98)
        -> maybe find_by_id (0.85)
  

• Incremental everywhere. Re-running tessera index . only re-parses files whose SHA changed. 951-file Java repo: full index 4.6 s, incremental rerun 64 ms.

• Memory-mapped graph snapshot. MCP server mmaps a bincode archive of the symbol + edge tables at startup. Hot-path queries don't touch SQLite.

• React-aware. .tsx parsed with the TSX grammar. <UserCard /> registers a reference to UserCard of kind jsx, so React component graphs work the same as call graphs.

• Call-path tracing. tessera connect handleRequest writeRow returns the shortest chain of calls from one symbol to another — "does A actually reach B, and how?" — a deterministic graph traversal an LLM-extracted graph can't answer reliably.

• Shareable graph export. tessera export --format mermaid (or dot) renders the call graph — the whole thing, or the precise forward subgraph rooted at a symbol with --from. Group by file, directory, or language; hide test nodes; filter to exported endpoints; or write a copy-button Mermaid preview with --html-out graph.html.

• Token-priced operations. Every response carries _meta with token estimates plus cheaper alternative queries. Agents can route to the right fidelity-to-token tradeoff.

A 30-second demo

tessera index examples/sample
tessera impact findById --json | jq '.callers[0] | {symbol: .symbol.qualified_name, criticality, breakdown}'

{
  "symbol": "renderUser",
  "criticality": 100.0,
  "breakdown": {
    "pagerank": 0.4674,
    "fanout_in": 0,
    "fanout_out": 1,
    "exported": true,
    "test_coverage": 0,
    "depth_decay": 1.0
  }
}

Wire it up to your coding agent

Tessera speaks MCP. Index your repo, point your agent at the binary.

Claude Code:

claude mcp add tessera tessera -- mcp --db .tessera/tessera.db

Cursor — add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):

{
  "mcpServers": {
    "tessera": {
      "command": "tessera",
      "args": ["mcp", "--db", "${workspaceFolder}/.tessera/tessera.db"]
    }
  }
}

Configs for Cline, Continue.dev, Codex CLI, Zed, Aider, and custom GPTs live in docs/integrations.md. Tool schemas in docs/mcp.md.

Exposed tools: find_definition · find_references · get_outline · expand_symbol · impact · connect · export · context_pack · plan_query · edit_prep · diff_impact · imports · imported_by · signature · siblings · search · unused · validate · validate_snippet · tests_for · stats.

Tip: run tessera doctor when a query looks stale or an MCP client cannot connect. It prints the exact tessera index . --db ... or tessera snapshot --db ... command to repair the local setup.

Start with the first five minutes guide, and read when not to use Tessera for the current precision limits. Troubleshooting lives in docs/troubleshooting.md.

Use as a Rust library

[dependencies]
tessera-codegraph = "0.4"

use tessera_codegraph::{Index, IndexOptions, Language};

let report = Index::build("./repo", "./repo/.tessera/tessera.db", IndexOptions::default())?;
let idx = Index::open("./repo/.tessera/tessera.db")?;

for caller in idx.impact("findById", 4)?.callers.iter().take(5) {
    println!("{:5.1}  {}", caller.criticality, caller.symbol.qualified_name);
}

let check = idx.validate_snippet("findByIdd(1)", Language::TypeScript)?;
println!("{} unresolved calls", check.unresolved_calls);

Languages

11 languages, each with real AST-level extraction (definitions, references, imports) — not a regex guess:

Language	Extensions	Notes
TypeScript	.ts, .mts, .cts	Parsed with the TSX grammar (a superset of TS)
TSX (React)	.tsx	<Component /> and <Foo.Bar /> register as references of kind jsx
JavaScript	.js, .jsx, .mjs, .cjs	JSX-aware; CommonJS require + ES6 import
Java	.java	Classes, interfaces, records, enums, methods, constructors, invocations, new
Python	.py, .pyw	Functions, classes
Go	.go	Functions, methods (receiver-qualified), structs, interfaces
Rust	.rs	Functions, methods, structs, enums, traits, modules, macro invocations
C	.c	Functions, structs, enums, unions, #include
C++	.cpp, .cc, .cxx, .h, .hpp, …	Classes, namespaces, methods, new; .h parsed as C++ (a C superset)
C#	.cs	Classes, interfaces, structs, records, enums, methods, using, new
Ruby	.rb	Classes, modules, methods, require
PHP	.php, .phtml	Classes, interfaces, traits, enums, functions, methods, use

More on the way (Kotlin, Swift, Scala, Lua, Zig — see ROADMAP.md). Skips .git, node_modules, target, dist, .next, .venv, __pycache__, and other common noise.

Reproduce the bench

tessera bench --path /path/to/your/repo
tessera bench --scale 200                # synthetic 200-file TS repo, no arguments
tessera bench --out docs/benchmarks.md   # write the chart to disk

The synthetic repo (tessera bench with no --path) models a "popular utility" topology: a sharedHelper called from every module file, mirroring how high-impact refactors really cascade through a codebase. See docs/benchmarks.md for methodology.

Architecture

Rust core · Tree-sitter (12 grammars, 11 languages) · SQLite (WAL, FTS5 trigram, Bloom) · memory-mapped snapshot · personalised PageRank impact · MCP stdio.

source files
  ─► tree-sitter parsers (ts/tsx/js, java, py, go, rust, c, c++, c#, ruby, php)
  ─► symbols, references, edges, exports
  ─► SQLite (WAL, FTS5 trigram, bloom blob)
  ─► memory-mapped snapshot (.tessera/snapshot.bin)
  ─► CLI · MCP server · library API

Details: docs/architecture.md · Quickstart: docs/quickstart.md · Benchmarks: docs/benchmarks.md

Development

cargo fmt --check
cargo clippy --all-targets --all-features -- -D warnings
cargo test --all-targets --all-features

Status

v0.9 — pre-alpha. 11 languages, JSX-aware React references, incremental indexing, PageRank-ranked impact, hallucination validator, agent workflow planning, call-path tracing (connect), shareable graph export (DOT/Mermaid/HTML preview), HTTP/SSE MCP transport with health metadata, and install via npm / Homebrew / curl / Docker / cargo — library + MCP + CLI. See CHANGELOG.md and ROADMAP.md.

Contributing

PRs welcome — parsers, graph accuracy, and query quality have the highest leverage. See CONTRIBUTING.md.

Sponsor

Tessera is Apache-2.0 and built in public. If it saves you or your team tokens, sponsor on GitHub to keep new languages, queries, and benchmarks shipping. Sponsors get early access to v0.4 features (runtime trace fusion, ADR memory, semantic git) and a say in the roadmap.

License

Apache-2.0 — see LICENSE.