SpecMem
The memory layer for your AI coding agents
SpecMem turns the specs, decisions, and tests scattered across your repo into searchable memory that any coding agent can use. One context layer, every agent.
Why SpecMem
Coding agents generate mountains of markdown, then forget all of it. SpecMem gives them a memory that lasts.
No vendor lock-in
Read specs from any agent format and serve them to any other. Switch between Kiro, Claude Code, Cursor, Codex, and more without rewriting a thing.
No more amnesia
A persistent, searchable memory of your specs, decisions, and tests. Agents recall earlier context instead of starting from scratch every session.
Less wasted compute
SpecImpact links specs, code, and tests, so agents run only what a change actually affects, and you see coverage gaps before they ship.
Everything in one layer
The building blocks that give your agents persistent, intelligent memory.
Multi-Framework Adapters
Parse specs from Kiro, SpecKit, Tessl, Claude Code, Cursor, Codex, Factory, Warp, and Gemini CLI.
Intelligent Memory
Vector-based semantic search with LanceDB, ChromaDB, Qdrant, or AgentVectorDB.
SpecImpact Graph
Bidirectional relationships between specs, code, and tests for instant impact analysis.
Coverage & Validation
Spot gaps between acceptance criteria and tests, with six built-in spec quality rules.
MCP Server
Model Context Protocol server so any agent can query specs and pull optimized context.
Web UI & Health Score
An interactive dashboard with live sync and an A to F project health grade.
Optimized Skills
Index tighter, easier-to-route versions of your agent skills, without ever touching the original files.
Source files stay untouched
Your SKILL.md files in .codex/skills and .claude/skills stay the source of truth. Optimized artifacts live separately.
A conservative quality gate
A candidate enters memory only when it improves on the source and passes static checks. Change the source skill, and the artifact is ignored until it is validated again.
Provenance you can trust
Optimized memory is tagged with its score change and artifact ID, so you always know which guidance an agent is reading and where it came from.
# Generate and gate an optimized candidate
specmem guidelines optimize .codex/skills/review/SKILL.md \
--instruction "tighten this for code review tasks"
# See which artifacts will be used
specmem guidelines optimized-status
# Build memory with accepted optimized skills
specmem build --optimize-skillsFully opt-in. A standard specmem build still indexes your source skills unchanged.
Up and running in minutes
Install it, point it at your specs, and query them from the CLI or Python.
Command line
pip install specmem[local]
cd your-project
specmem init # create .specmem.toml
specmem scan # detect and index your specs
specmem serve # open the dashboard at :8765Python API
from specmem import SpecMemClient
sm = SpecMemClient()
sm.index_specs()
# Ask what constrains a change
results = sm.search("authentication")
impact = sm.analyze_impact("src/auth.py")
coverage = sm.get_coverage()Prefer to see it on a real repo?
Clone our example project and ask it a plain-English question. You get back the specs, decisions, and tests that constrain the change you are about to make, ranked by relevance.
git clone https://github.com/SuperagenticAI/specmem-demo && cd specmem-demo
uv sync
uv run specmem scan . && uv run specmem build .
uv run specmem query "What should I know before changing src/auth/service.py?" -k 5Coding agent support
Agent-agnostic by design. SpecMem reads every major spec framework and AI coding agent.
Spec frameworks
| Framework | Adapter | Patterns |
|---|---|---|
| Kiro | kiro | .kiro/specs/**/*.md |
| SpecKit | speckit | .speckit/**/*.yaml |
| Tessl | tessl | .tessl/**/*.md |
Commercial agents
| Agent | Adapter | Patterns |
|---|---|---|
| Claude Code | claude | Claude.md, CLAUDE.md |
| Cursor | cursor | cursor.json, .cursorrules |
| Codex | codex | .codex/**/*.md |
| Factory | factory | .factory/**/*.yaml |
| Warp | warp | .warp/**/*.md |
| Gemini CLI | gemini | GEMINI.md, .gemini/**/*.md |
Database & embedding support
Bring your own vector store and embedding model. Run fully local or plug into the cloud.
Vector databases
LanceDB
Local development, default
pip install specmem[local]ChromaDB
Embedded persistence
pip install specmem[chroma]Qdrant
Production scale
pip install specmem[qdrant]AgentVectorDB
Lightweight, in-memory
pip install Built-inEmbedding models
OpenAI
text-embedding-3-small
pip install specmem[openai]embedding-001
pip install specmem[google]Together AI
togethercomputer/m2-bert
pip install specmem[together]Local
all-MiniLM-L6-v2
pip install specmem[local]Advanced features
Power tools for teams and production pipelines.
SpecDiff Timeline
Track spec evolution, detect drift, and find contradictions over time.
Test Mapping Engine
Map tests to specs across pytest, jest, vitest, and more.
Coding Guidelines
Aggregate guidelines from Kiro steering, CLAUDE.md, and .cursorrules.
Spec Lifecycle
Prune stale specs, generate specs from code, and compress verbose ones.
Kiro Session Search
Index and search Kiro chat sessions for extra project context.
Static Dashboard
Export your specs to a static HTML dashboard for GitHub Pages.
GitHub Action
CI integration with coverage thresholds and automatic PR comments.
Streaming Context API
Token-aware context optimization with streaming support.
Give your agents memory
Open source and Apache-2.0 licensed. Start with a single command.