Build AI agents
nobody can take away.

Kestrel is a production-ready framework for AI agents with cryptographic identity, persistent memory you own, and constitutional governance enforced above the LLM. Not another wrapper. Infrastructure.

Get early access View on GitHub

700+

issues shipped

sovereignty pillars

4,000+

unit tests

Apache 2.0

open source license

Architecture

Three pillars of sovereignty

Sovereignty isn't a feature you add on top. It's a property that has to be designed in from the start — across identity, memory, and governance.

01 / Identity

Portable DID identity

Every agent is born with a did:pkh decentralized identifier — a cryptographic keypair using the same secp256k1 curve as Ethereum. The agent's identity is not issued by any platform. It's generated locally, owned by the user, and portable to any environment.

did:pkh:eip155:1:0x8955b8cB...

02 / Memory

Persistent memory you own

Conversations, documents, and relationships are stored in a local SQLite knowledge graph with full-text search and RAG. The memory is yours — encrypted at rest, portable, and deletable on demand. The agent's memory doesn't live in our cloud.

SQLite · FTS5 · Encrypted at rest

03 / Governance

Constitutional governance

Every agent runs under an audited constitution enforced above the LLM — not inside the model, not in a prompt. Genesis audit on creation. Amendment requires cryptographic signature. The model can't override it. Neither can a bad prompt.

Genesis hash · Amendment log · Audit trail

See it in action

A sovereign agent, end to end

Every screenshot is the real Sovereign Console running on a machine you control — identity, governance, memory, privacy, and permissions, with nothing hidden in a cloud you don’t own.

Identity panel showing the agent's did:pkh decentralized identifier — **Portable DID identity.** Born with a `did:pkh` keypair the agent owns — not issued by any platform.

Constitution panel with the Kestrel Constitution and its verified hash — **Constitutional governance.** A hashed, tamper-evident constitution checked on every interaction.

Knowledge graph panel showing typed memory nodes — **Persistent memory you own.** A local knowledge graph — encrypted, portable, deletable on demand.

Chat header in EPHEMERAL privacy mode using a local Ollama model — **Privacy as a spectrum.** EPHEMERAL routes to a local model only; NORMAL can use the cloud — you choose, per agent.

Sovereignty export result — a portable snapshot of the agent — **Export everything, anytime.** DID, constitution, memory, and history in one portable snapshot — restore it anywhere.

Security panel showing a blocked action after a permission is denied — **Permission enforcement.** Deny a capability and the agent is blocked — with an audit trail of every attempt.

Developer Experience

Running in minutes,
not weeks

One command to create an agent. Another to start it. Your agent is up with a full web UI, DID identity, memory store, and constitutional protections baked in.

Read the quickstart →

terminal

# Install and set up
git clone github.com/KestrelSovereignAI/kestrel-sovereign
cd kestrel-sovereign
uv sync

# Create your agent — gets a DID, memory store,
# constitution, and web UI automatically
uv run kestrel create MyAgent

# Start it
uv run kestrel start MyAgent

# → Agent running at http://localhost:8888
# → DID: did:pkh:eip155:1:0x...
# → Constitution hash: 0xf3a2...
# → Memory store: encrypted, local, yours
        

Open Source

The foundation is open.
The ownership stays with you.

Kestrel is Apache 2.0 licensed. The sovereign core — identity, memory, and constitutional governance — is fully open. No lock-in by design.

⚖️

Apache 2.0 Licensed

Use it, fork it, build on it. No hidden restrictions on commercial use or modification.

🔒

No Platform Lock-In

Your agent's identity and memory are not hosted by us. Self-host anywhere — your laptop, your server, your cloud.

🔌

Multi-LLM Support

Anthropic, OpenAI, Gemini, Ollama — run fully local. The framework is provider-agnostic by design.

🐍

Python · FastAPI · uv

Modern Python stack with a clean install story. One command to set up, runs on Mac, Linux, and Windows.

Blog

Teach the market what sovereign AI means

Kestrel needs a place to explain the why behind sovereign AI, not just the code. Start with a few clear essays. Grow it into a real publication later.

Visit the blog →

Why Kestrel 5 min read

What Sovereign AI Actually Means

A plain-language explanation of why identity, memory, and governance belong to the user, not the platform.

Read article →

Memory Ownership 6 min read

Why Agent Memory Should Belong to the User

If your agent cannot carry its memory with it, you do not really own the relationship or the context it builds.

Read article →

Governance 7 min read

Constitutional Governance for Agents

Why model behavior cannot be trusted to prompting alone when long-term, user-owned agents are making important decisions.