v0.3.5-beta.2 · BSL 1.1

ThreadWeaver

AI chat where every conversation is versioned, branchable, and searchable — powered by AgentStateGraph.

SvelteKit + FastAPI + Rust · Self-host in 60 seconds

View on GitHub Read the Docs Self-Host Now

Not another chat wrapper

Branch any conversation

Every message is a branch point. Click to fork and explore "what if I'd asked differently?" without losing the original thread. Branches are O(1) — instant creation, instant discard. Compare approaches side by side, then commit the winner.

You
Assistant
Branch point
Original thread
3 messages
What if I asked differently?
2 messages

Full provenance on every message

Every message exchange is a versioned commit with intent, reasoning, and confidence metadata. Query by "why," not just "what." Blame any value to see who changed it, when, and with what reasoning. Powered by AgentStateGraph.

Checkpoint2:31 PM
Initialize cluster configuration
agent/setup95%
Explore2:32 PM
Evaluate network topology options
agent/planner72%
Refine2:33 PM
Apply mesh topology with failover
agent/planner89%

Rich markdown rendering

Code blocks with syntax highlighting, tables, headings, lists, links, blockquotes — all rendered inline. Paste images, drop files. Works with Claude, GPT, Gemini, Grok, and local models via Ollama.

Assistant
Setting up the pipeline
async def deploy(config):
  # Validate and push
  await validate(config)
  return push("prod")

Tool calling + MCP servers

Connect MCP servers with one click — filesystem, GitHub, SQLite, web search, and 100+ more. The AI reads files, runs commands, and queries databases. Full agentic loop with automatic tool dispatch and result integration.

You
List the files in src/
Assistant
list_directory
src/lib/
src/routes/
src/app.html
src/app.css
Found 4 entries in src/. The project has a standard SvelteKit structure.

Works with your models

Claude GPT Gemini Grok Ollama OpenAI-compatible

Three layers, one primitive

SvelteKit Frontend
FastAPI Backend
AgentStateGraph (Rust)
LLM Providers
Claude GPT Gemini Grok Ollama Custom
MCP Servers
Filesystem GitHub SQLite Memory Search 100+ more

What makes ThreadWeaver different

FeatureTypical AI ChatThreadWeaver
Branching✗ Linear scroll only✓ Fork from any message
📊Provenance✗ No history metadata✓ Intent, reasoning, confidence per commit
🔎Search✗ Ctrl+F at best✓ Full-text across all conversations
📓Notebook✗ Copy-paste✓ Tag, comment, branch from highlights
🛠Tool calling✗ Chat only✓ MCP servers + agentic loop
📁Projects✗ Flat list✓ Group related chats with shared context
🎨Multimodal✓ Some providers✓ All 6 providers + paste/drop
🌐Self-hosted✗ Cloud only✓ Docker, ARM64, your data stays yours
📱BYOK✗ Subscription required✓ Bring your own API keys

Self-host in 60 seconds

terminal

      

Docker

docker compose up -d

Backend + frontend, one command

+ Ollama

docker compose -f ... up -d

Add local models (Llama, Mistral, Qwen)

Native

pip install && npm run dev

ARM64 / Jetson / PicoClaw ready

Built on AgentStateGraph

  • Every message exchange is a versioned commit with intent, reasoning, and confidence metadata
  • Sealed epochs provide tamper-evident audit trails — prove what happened, when, and why
  • Branch a conversation and the entire state graph forks — O(1) creation, instant discard
  • Authority chains track who authorized what, with full delegation history
a7f3...c2
commit
d1e8...9a
commit
f5b2...17
commit
SEALED

Tamper-evident epoch — every commit hashed, the chain verifiable offline