haft vs IntelliCode
Side-by-side comparison to help you choose.
| Feature | haft | IntelliCode |
|---|---|---|
| Type | Agent | Extension |
| UnfragileRank | 45/100 | 40/100 |
| Adoption | 0 | 1 |
| Quality | 0 | 0 |
| Ecosystem | 1 |
| 0 |
| Match Graph | 0 | 0 |
| Pricing | Free | Free |
| Capabilities | 11 decomposed | 7 decomposed |
| Times Matched | 0 | 0 |
Enforces a disciplined 5-mode engineering cycle (Understand → Explore → Choose → Execute → Verify) by requiring AI agents to frame problems before solving them, generate genuinely different solution variants, and compare options under parity constraints. Implements this via MCP tools (haft_problem, haft_solution, haft_decision) that validate reasoning artifacts against a formal specification before allowing progression to implementation.
Unique: Implements a formal specification-driven reasoning cycle with maturity (Unassessed → Shipped) and freshness (Healthy → Stale → At Risk) tracking, enforcing parity in comparisons via a knowledge graph that links decisions to codebase artifacts — unlike generic prompt engineering, this creates falsifiable contracts with evidence decay mechanics
vs alternatives: Differs from Cursor/Claude Code's native reasoning by adding governance layer that prevents decision drift and enforces structured comparison, whereas standard agents optimize for speed-to-code
Exposes Haft's reasoning capabilities as a Model Context Protocol (MCP) server via JSON-RPC transport, providing six specialized tools (haft_problem, haft_solution, haft_decision, haft_evidence, haft_check, haft_search) that AI agents can invoke natively within their execution environment. The server runs as a subprocess managed by the agent's MCP client, maintaining a persistent SQLite state store and knowledge graph indexed to the codebase.
Unique: Implements MCP as the primary delivery surface (not a secondary plugin), with six domain-specific tools designed for the FPF cycle rather than generic function calling — includes codebase-aware search and evidence decay scoring built into the protocol layer
vs alternatives: More specialized than generic MCP servers (e.g., Anthropic's file-system MCP) because tools are designed for reasoning governance, not file I/O; tighter integration with decision lifecycle than REST APIs
Enforces equal rigor in comparing competing solutions by requiring that all variants be evaluated against the same criteria, preventing bias toward preferred solutions. Implements parity checks via the haft_solution and haft_decision tools that validate solution descriptions follow the same structure and depth. Tracks comparison fairness metrics to ensure decisions are based on equivalent evidence.
Unique: Implements structural parity checks that validate all solutions follow the same evaluation template and depth — unlike generic decision frameworks, this prevents strawman alternatives and ensures fair comparison
vs alternatives: More rigorous than informal decision-making because it enforces structural equivalence; differs from decision matrices by focusing on comparison process rather than scoring
Monitors the health of engineering decisions across two axes: maturity (progress from Unassessed to Shipped) and freshness (Healthy → Stale → At Risk based on evidence age and drift detection). Implements R_eff (effective reasoning score) that decays over time as supporting evidence ages, triggering alerts when decisions drift from their original context. Uses SQLite schema with timestamp-based queries to identify stale decisions and prompt re-evaluation.
Unique: Implements a two-axis decision lifecycle model (maturity + freshness) with time-decay scoring (R_eff) that automatically degrades decision confidence — unlike static decision logs, this creates a living system where old decisions are flagged for re-evaluation without manual intervention
vs alternatives: More sophisticated than ADR (Architecture Decision Records) because it tracks decision health over time and flags staleness; differs from code review tools by focusing on decision validity rather than code quality
Builds a knowledge graph that links engineering decisions to codebase artifacts (modules, functions, files) using FPF Spec Search & Indexer. Enables semantic search over past decisions filtered by codebase context, allowing agents to query 'decisions affecting this module' or 'solutions tried for this problem pattern'. Stores graph in SQLite with projections that map decisions to code locations and vice versa.
Unique: Implements a bidirectional knowledge graph (decisions ↔ code artifacts) with FPF Spec Search that understands decision semantics and codebase structure simultaneously — unlike generic code search, this links reasoning to implementation and enables decision-centric queries
vs alternatives: More targeted than full-text search because it understands decision structure and codebase topology; differs from RAG systems by maintaining explicit decision-to-code mappings rather than embedding-based retrieval
Provides a terminal-based autonomous agent (haft agent command) that executes the engineering cycle without human intervention, using a ReAct-style coordinator to move through Understand → Explore → Choose → Execute → Verify phases. The coordinator maintains state in SQLite and can pause at checkpoints for human review. Implements a lemniscate cycle pattern that allows looping back to earlier phases if verification fails.
Unique: Implements a lemniscate cycle (figure-8 loop) that allows backtracking from Verify to earlier phases if verification fails, rather than linear progression — enables iterative refinement without restarting the entire cycle
vs alternatives: More structured than generic ReAct agents because it enforces FPF phases; differs from Devin/Claude Code by running autonomously in terminal without IDE, making it suitable for headless environments
Abstracts LLM provider differences (OpenAI Codex, Anthropic Claude, Google Gemini) behind a unified interface, allowing the same FPF reasoning cycle to work across different models. Routes tool calls and reasoning prompts to the configured provider via a provider adapter pattern, with fallback support for multiple models. Stores provider configuration in project policy files.
Unique: Implements provider abstraction at the reasoning level (not just API calls), allowing the same FPF cycle to work across Claude, Codex, and Gemini with different tool-calling conventions — uses adapter pattern to normalize provider differences
vs alternatives: More flexible than single-provider agents (Claude Code, Cursor) because it supports provider switching; differs from LangChain by focusing on reasoning governance rather than generic LLM chaining
Enforces project-level governance policies via .haft/ directory containing formal specifications (FPF Spec), provider configurations, and decision templates. Policies are versioned and can be checked via haft check command to ensure decisions comply with project standards. Implements a policy-as-code approach where governance rules are stored alongside the project and enforced by the Haft runtime.
Unique: Implements governance as versioned policy files in .haft/ directory (similar to .github/ workflows), making policies auditable and version-controlled alongside code — unlike external governance systems, policies live in the repository
vs alternatives: More integrated than external compliance tools because policies are co-located with code; differs from linters by enforcing reasoning discipline rather than code style
+3 more capabilities
Provides IntelliSense completions ranked by a machine learning model trained on patterns from thousands of open-source repositories. The model learns which completions are most contextually relevant based on code patterns, variable names, and surrounding context, surfacing the most probable next token with a star indicator in the VS Code completion menu. This differs from simple frequency-based ranking by incorporating semantic understanding of code context.
Unique: Uses a neural model trained on open-source repository patterns to rank completions by likelihood rather than simple frequency or alphabetical ordering; the star indicator explicitly surfaces the top recommendation, making it discoverable without scrolling
vs alternatives: Faster than Copilot for single-token completions because it leverages lightweight ranking rather than full generative inference, and more transparent than generic IntelliSense because starred recommendations are explicitly marked
Ingests and learns from patterns across thousands of open-source repositories across Python, TypeScript, JavaScript, and Java to build a statistical model of common code patterns, API usage, and naming conventions. This model is baked into the extension and used to contextualize all completion suggestions. The learning happens offline during model training; the extension itself consumes the pre-trained model without further learning from user code.
Unique: Explicitly trained on thousands of public repositories to extract statistical patterns of idiomatic code; this training is transparent (Microsoft publishes which repos are included) and the model is frozen at extension release time, ensuring reproducibility and auditability
vs alternatives: More transparent than proprietary models because training data sources are disclosed; more focused on pattern matching than Copilot, which generates novel code, making it lighter-weight and faster for completion ranking
haft scores higher at 45/100 vs IntelliCode at 40/100. haft leads on quality and ecosystem, while IntelliCode is stronger on adoption.
Need something different?
Search the match graph →© 2026 Unfragile. Stronger through disorder.
Analyzes the immediate code context (variable names, function signatures, imported modules, class scope) to rank completions contextually rather than globally. The model considers what symbols are in scope, what types are expected, and what the surrounding code is doing to adjust the ranking of suggestions. This is implemented by passing a window of surrounding code (typically 50-200 tokens) to the inference model along with the completion request.
Unique: Incorporates local code context (variable names, types, scope) into the ranking model rather than treating each completion request in isolation; this is done by passing a fixed-size context window to the neural model, enabling scope-aware ranking without full semantic analysis
vs alternatives: More accurate than frequency-based ranking because it considers what's in scope; lighter-weight than full type inference because it uses syntactic context and learned patterns rather than building a complete type graph
Integrates ranked completions directly into VS Code's native IntelliSense menu by adding a star (★) indicator next to the top-ranked suggestion. This is implemented as a custom completion item provider that hooks into VS Code's CompletionItemProvider API, allowing IntelliCode to inject its ranked suggestions alongside built-in language server completions. The star is a visual affordance that makes the recommendation discoverable without requiring the user to change their completion workflow.
Unique: Uses VS Code's CompletionItemProvider API to inject ranked suggestions directly into the native IntelliSense menu with a star indicator, avoiding the need for a separate UI panel or modal and keeping the completion workflow unchanged
vs alternatives: More seamless than Copilot's separate suggestion panel because it integrates into the existing IntelliSense menu; more discoverable than silent ranking because the star makes the recommendation explicit
Maintains separate, language-specific neural models trained on repositories in each supported language (Python, TypeScript, JavaScript, Java). Each model is optimized for the syntax, idioms, and common patterns of its language. The extension detects the file language and routes completion requests to the appropriate model. This allows for more accurate recommendations than a single multi-language model because each model learns language-specific patterns.
Unique: Trains and deploys separate neural models per language rather than a single multi-language model, allowing each model to specialize in language-specific syntax, idioms, and conventions; this is more complex to maintain but produces more accurate recommendations than a generalist approach
vs alternatives: More accurate than single-model approaches like Copilot's base model because each language model is optimized for its domain; more maintainable than rule-based systems because patterns are learned rather than hand-coded
Executes the completion ranking model on Microsoft's servers rather than locally on the user's machine. When a completion request is triggered, the extension sends the code context and cursor position to Microsoft's inference service, which runs the model and returns ranked suggestions. This approach allows for larger, more sophisticated models than would be practical to ship with the extension, and enables model updates without requiring users to download new extension versions.
Unique: Offloads model inference to Microsoft's cloud infrastructure rather than running locally, enabling larger models and automatic updates but requiring internet connectivity and accepting privacy tradeoffs of sending code context to external servers
vs alternatives: More sophisticated models than local approaches because server-side inference can use larger, slower models; more convenient than self-hosted solutions because no infrastructure setup is required, but less private than local-only alternatives
Learns and recommends common API and library usage patterns from open-source repositories. When a developer starts typing a method call or API usage, the model ranks suggestions based on how that API is typically used in the training data. For example, if a developer types `requests.get(`, the model will rank common parameters like `url=` and `timeout=` based on frequency in the training corpus. This is implemented by training the model on API call sequences and parameter patterns extracted from the training repositories.
Unique: Extracts and learns API usage patterns (parameter names, method chains, common argument values) from open-source repositories, allowing the model to recommend not just what methods exist but how they are typically used in practice
vs alternatives: More practical than static documentation because it shows real-world usage patterns; more accurate than generic completion because it ranks by actual usage frequency in the training data