Key takeaways
- Graphify connects code, documentation, and other project material in a graph with explicit provenance labels.
- Code-only extraction is local; document and media semantics can use a configured model provider.
- The CLI is Apache-2.0, while hosted plans have separate limits and subscription terms.
- Inference labels and graph paths help inspection but do not prove every dependency is correct.
FAQ
What is Graphify?
A code and document graph engine that coding assistants can query through a CLI, agent skill, or MCP server.
Does Graphify require a model API?
Code-only AST extraction does not. Semantic processing of documents and media uses the assistant's model or a configured backend, which can be local.
How much does Graphify cost?
The local CLI is free under Apache-2.0. The hosted service lists a free tier and Pro at $15 monthly or $120 annually; team and enterprise plans are separate.
Is Graphify's graph definitive?
No. Edges distinguish explicit extraction, inference, and ambiguity, and language-resolution or incremental-update errors can still affect results.
Executive Summary
Graphify maps source code and related project material into a persistent graph that an AI assistant can query. It qualifies for the code intelligence tools comparison through shipped code extraction, relationship queries, and agent integrations; its document and media processing extends beyond that category.[1]
The key distinction is between deterministic parsing and model-assisted interpretation. Code-only extraction runs locally. Documents and media can be sent to an assistant or configured model backend. Those paths should not be covered by a blanket offline claim.[1]
| Attribute | Checked September 15, 2026 |
|---|---|
| Attribution | Safi Shamsi and Graphify contributors[2] |
| Current release | v0.9.62[3] |
| Local license | Apache-2.0; older MIT-contributed portions retain their terms[2] |
| Hosted product | Free, Pro, Teams, and self-hosted Enterprise plans[4] |
Product Overview
Install the graphifyy Python package, then use graphify install to register the assistant skill. The generated artifacts include a graph JSON file, an interactive HTML view, and a Markdown report.[1]
| Surface | Purpose | Availability |
|---|---|---|
| CLI and assistant skill | Build, query, explain, and traverse project graphs | Public package[1] |
| MCP server | Repeated queries through stdio or optional HTTP transport | Documented installation[1] |
| Hosted service | Managed graphs and review workflows | Public plans; check operational limits[4] |
Technical Architecture
The Python library passes extracted nodes and edges through graph construction, community detection, analysis, and export modules. It uses NetworkX graph structures and supports JSON, HTML, GraphML, and other outputs. This is inspectable graph state rather than a required vector-database deployment.[5]
The schema distinguishes EXTRACTED relationships explicitly present in source, INFERRED relationships derived by resolution, and AMBIGUOUS relationships flagged for review. Source paths and locations accompany nodes. These labels expose how a relationship was obtained; they are not calibrated probabilities or correctness proofs.[5]
Recent release work is concrete: v0.9.62 adds local Terraform module topology, conservative inherited Ruby-call resolution, and unambiguous Markdown code references. It also fixes several false-edge and incremental-write problems. Some dynamic or ambiguous cases deliberately remain unresolved.[3]
Strengths
- Mixed project context: documentation references can connect to actual code symbols, useful when design rationale lives outside the source tree.[3]
- Visible provenance: edge labels help reviewers distinguish explicit relationships from deductions.[5]
- Deployment choice: the local engine and optional served graph can be evaluated independently of hosted subscriptions.[1][4]
Cautions
- Model boundaries: semantic document processing can incur model charges and send content to the chosen provider.[1]
- Documentation drift: the security policy still lists 0.3.x as supported while the current release is 0.9.62. Its broad network wording should be read alongside the README's explicit model-backend behavior.[6][3]
- Untrusted content: the policy describes prompt-injection defenses but explicitly says they cannot make injection impossible.[6]
- Index integrity: an open incremental-update report merits testing on repositories with submodules and untracked files.[7]
What Developers Say
On September 15, GitHub user turzail-png reported that v0.9.42 on Windows lost graph nodes for 168 unchanged files after an incremental post-commit rebuild. Most affected files were in submodules or untracked paths; a full rebuild reportedly restored them. The issue remained open at review time. It is a useful first-hand failure report on an older release, not a reproduced finding about v0.9.62.[7]
This review did not independently run Graphify or validate its promotional token-savings claims. The recommendation therefore rests on documented functionality and identifiable evaluation risks.
Pricing & Licensing
Prices below are USD, checked September 15, 2026. Hosted subscriptions are separate from the free local engine.[4]
| Offering | Published price | Main constraint |
|---|---|---|
| Local CLI | Free | Apache-2.0 and retained notices[2] |
| Hosted Free | $0 | 25,000 nodes/repo, 10 push builds/repo/day, 15 reviews and verification runs/month |
| Hosted Pro | $15/month or $120/year | One developer; shared operational limits apply |
| Hosted Teams | $29/seat/month or $240/seat/year initially | First 100 teams; then $40/month or $336/year per seat; minimum two seats |
| Enterprise | Custom | Self-hosted, negotiated scope |
“Uncapped” paid features still have concurrency, rate, and memory limits. Hosted trials require a card; the free tier does not. Local semantic extraction can add model-provider costs.[4][1]
Competitive Positioning
Within code intelligence, Graphify is worth evaluating when code must connect to documentation and other artifacts, and when inspectable relationship provenance matters. Its public architecture makes those evaluation criteria concrete.[5]
Ideal Customer Profile
Best fit: developers exploring a repository alongside its design documents, with time to test graph coverage and choose the model/deployment boundary.
Poor fit: teams treating every inferred edge as authoritative or expecting the hosted and local products to have identical privacy and billing behavior.
Viability Assessment
A current release and detailed public implementation provide evidence of active development. They do not establish enterprise reliability, financial durability, or independent performance. The release's conservative resolution changes are more informative than a popularity ranking.[3]
Bottom Line
Graphify offers a practical way to connect project structure with surrounding knowledge. Evaluate it by checking meaningful paths, missing edges, update behavior, and the costs of semantic processing.
Recommended for: mixed code-and-document exploration.
Not recommended for: treating generated graphs as proof of program behavior.
Outlook: useful breadth, with indexing accuracy and deployment boundaries requiring continued scrutiny.
Research by Ry Walker Research • methodology
Sources
- [1] Graphify README — installation, integrations, and privacy
- [2] Graphify licensing notice
- [3] Graphify v0.9.62 release — September 15, 2026
- [4] Graphify hosted pricing
- [5] Graphify architecture and graph schema
- [6] Graphify security policy
- [7] Graphify issue #3580 — incremental graph loss report on v0.9.42