kestrel-sovereign 0.27.0

Kestrel Sovereign AI Agent Framework - Constitutional AI with cryptographic identity

Kestrel: Sovereign AI Agent Framework

Build AI agents that nobody can take away from their users — not you, not the cloud, not the next pivot.

Kestrel is a continuously-developing framework for creating autonomous AI agents with cryptographic identity, persistent memory, and constitutional governance. Every agent you deploy is owned by its user, governed by immutable principles, and able to remember across every conversation. The core install is stable enough to run real agents today; the surrounding ecosystem (cloud providers, training adapters, integrations) is actively evolving — see Feature Stability for the current per-feature picture.

Three Pillars

Pillar	What it means
Portable DID identity	Cryptographic identity the agent's user owns. Exportable, self-hostable, cloud-optional — the agent is not bound to any provider.
Persistent memory you own	Local-first memory with full-text search, knowledge graph retrieval, and RAG. Conversations, documents, relationships — searchable, portable, and encrypted at rest when configured. SQLAlchemy-backed vector storage is in tree for saved items, document chunks, and conversation history; embedding generation is still being standardized across LLM providers.
Constitutional governance	Every agent runs under an audited set of principles enforced above the LLM. Genesis audit on creation. Amendment requires cryptographic signature.

What's in core, what's an add-on

pip install kestrel-sovereign gives you a complete, working sovereign agent: identity, memory, constitution, privacy modes, multi-LLM support, local guarded compute, and a Cloud Run deployment path. Everything you need to run an agent locally with zero cloud commitment.

Voice, MCP, GitHub App, wallet, council, observability, and similar specialized capabilities are installable feature packages. RunPod, Vast.ai, GCP Compute, voice cloud backends, and storage backends are provider packages that register with provider-specific entry points rather than the feature entry-point group. This split is being completed across #462 and #560; current state is documented in KESTREL_FEATURES.md.

🚀 Quick Start

Prerequisites

• Python 3.11-3.14
• uv (for package management)
• Ollama (optional - for local LLM inference without API keys)

Install uv

If you don't have uv installed:

# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# Or with pip
pip install uv

Installation

# 1. Clone and setup
git clone https://github.com/KestrelSovereignAI/kestrel-sovereign.git
cd kestrel-sovereign
uv sync  # Creates .venv and installs all dependencies

# 2. (Optional) Start Ollama for local models - skip if using cloud APIs
ollama serve
ollama pull llama3.2:3b

# 3. Run the setup wizard. Interactive by default; --quickstart
#    accepts every default non-interactively. --quickstart auto-detects
#    available LLM providers — checks for cloud API keys
#    (OPENROUTER_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY,
#    GOOGLE_API_KEY) in your shell env, probes Ollama at
#    localhost:11434, and writes the priority list accordingly. If
#    none are available, it falls back to Ollama-only.
uv run kestrel setup               # interactive
uv run kestrel setup --quickstart  # non-interactive; auto-detects providers
# Or hand-edit: cp kestrel.toml.example kestrel.toml

# 4. Doctor check (verify readiness)
uv run kestrel doctor

# 5. (Optional) Create an additional agent. `--quickstart` already
#    registers one named `Kestrel`; this step is for adding more or
#    if you ran the interactive wizard without auto-registering.
uv run kestrel create MyAgent

# 6. Start an agent. Name the one you want, or omit to start every
#    agent registered with autostart=true (the wizard's default).
#    After --quickstart with no step 5, the autostart agent is "Kestrel".
uv run kestrel start            # starts every agent with autostart=true
uv run kestrel start Kestrel    # if you only ran --quickstart
uv run kestrel start MyAgent    # if you ran step 5 (works regardless of autostart)

If you're upgrading from a pre-2026-05 setup that used a standalone llm_config.toml, run uv run kestrel migrate-llm-config to fold it into kestrel.toml [llm]. The legacy file is no longer read.

Your agent is now running. Two ports to know about, depending on which start form you used:

Command	Listens on	Why
kestrel start (no name)	http://localhost:8888 (the multi-agent host)	Default in-process multi-agent mode; the host fronts every agent registered with autostart = true at multi_agent.host.port (default 8888). Agents registered with autostart = false aren't loaded — start those by name.
kestrel start <name>	the agent's own port, printed by the CLI on start	Each agent gets the next free slot at or above 8801 (the first auto-assigned agent lands there; subsequent agents go to 8802, etc., or whatever --port you passed to kestrel create). The exact value lives in multi_agent.toml under that agent's entry, and kestrel start <name> prints Starting <name> on :<port>....

Port conflict? Edit the agent's entry in multi_agent.toml to change its port, or recreate the agent with a chosen port (kestrel create MyAgent --port 8899). Edit multi_agent.toml's [host] section to change the host port (default 8888). kestrel start itself doesn't take a --port flag — runtime ports are read from multi_agent.toml.

Test it: Visit the URL the CLI printed on start (http://localhost:8888 for the multi-agent host, or whatever per-agent port kestrel start <name> reported). The Sovereign Console is the default page; append /health for a JSON readiness probe.

Windows users: the CLI prints emoji. If you see UnicodeEncodeError: 'charmap' codec can't encode character ..., run chcp 65001 once in your PowerShell session to switch the console to UTF-8. (As of v0.1.9 the CLI auto-reconfigures stdout, so a fresh install should not hit this.)

Install from PyPI (no source clone)

The Quick Start above clones the repo so you have demos, examples, and the kestrel.toml.example next to your agent. If you only want to run an agent and don't need the source tree, install from PyPI:

# 1. Install the CLI (uv tool install is preferred — `kestrel`
#    lands on PATH in an isolated venv. Plain `pip install
#    kestrel-sovereign` works too, into whichever venv is active.)
uv tool install kestrel-sovereign

# 2. Pick where Kestrel keeps your data. Either set KESTREL_HOME
#    explicitly (recommended for ops) — or skip this and Kestrel will
#    use ~/.kestrel/ by default.
export KESTREL_HOME="$HOME/kestrel-data"

# 3. Same wizard as above. Writes .env, kestrel.toml, multi_agent.toml,
#    and agent_data/ under $KESTREL_HOME.
kestrel setup --quickstart
kestrel start

Where data lives. kestrel resolves the project directory in this order: KESTREL_HOME → walk up from CWD looking for a multi_agent.toml / kestrel.toml / .env marker → ~/.kestrel/ for pip-installed users with no markers anywhere. A pure pip install with no KESTREL_HOME and no project in CWD lands on ~/.kestrel/ and creates it on first run. Never writes to site-packages/ — pip install --upgrade kestrel-sovereign is safe and won't touch your agent data.

If you later want to switch your data dir, move it: mv ~/.kestrel /new/path && export KESTREL_HOME=/new/path. The agent's encrypted DB is portable; nothing about the data dir is hard-coded.

CLI Commands (Cross-Platform)

All commands work on Windows, macOS, and Linux. Pass the agent directory as an argument:

Running kestrel directly. The commands below are written as bare kestrel …. That works whenever the kestrel console script is on your PATH — i.e. your project venv is activated (source .venv/bin/activate, or .venv\Scripts\activate on Windows) or you installed it globally with uv tool install. Working from a source clone without activating? Prefix any command with uv run (e.g. uv run kestrel status) and uv will resolve the project environment for you.

kestrel health                       # Check prerequisites
kestrel create MyAgent               # Create a new agent
kestrel start MyAgent                # Start an agent
kestrel stop MyAgent                 # Stop an agent
kestrel restart MyAgent              # Restart (stop then start)
kestrel update [MyAgent]             # Pull + install + feature sync + restart (see below)
kestrel status                       # Show all running agents
kestrel list                         # List available agents
kestrel shell MyAgent                # CLI chat interface
kestrel config ./agent_data/MyAgent  # Show agent config

Pulling in upstream changes (kestrel update)

The typical refresh-the-checkout loop is four commands:

git pull                               # new code from origin
uv sync                                # refresh deps from uv.lock (PRUNES kestrel-feature-* packages)
kestrel feature sync                   # restore the feature packages uv sync just pruned
kestrel restart                        # bring agents back so they see the new install

kestrel update collapses that into one verb:

kestrel update                       # pull + install + feature sync + restart all agents
kestrel update Emma                  # same, but only restart the named agent
kestrel update --dry-run             # preview the steps without mutating anything
kestrel update --no-pull --no-install # only feature sync + restart (e.g. after a manual checkout)
kestrel update --allow-dirty         # pull even with modified TRACKED files in the working tree
kestrel update --uv-sync             # force `uv sync` for the install step
kestrel update --no-uv-sync          # force `uv pip install -e .` for the install step
kestrel update --no-deps             # pass --no-deps to `uv pip install` for the fast path
kestrel update --continue-on-error   # restart even if `feature sync` reports an error

Install step auto-detect. When uv.lock exists at the source checkout root, the install step runs uv sync --active against the venv that owns the running kestrel binary (detected via sys.prefix != sys.base_prefix, so systemd / cron / direct-path invocations work too). That refreshes deps from the lockfile AND prunes packages not in it — which is precisely why kestrel feature sync runs immediately after, to restore any out-of-tree feature packages (kestrel-feature-*) that uv sync just pruned. Without uv.lock present, the step falls back to uv pip install --python sys.executable -e . which only reinstalls the project package. Force either branch with --uv-sync / --no-uv-sync.

Dirty-tree detection. Only modified or staged tracked files block the pull. Untracked files (kestrel.toml.backup-* written by kestrel setup, scratch files, etc.) don't count — git pull --ff-only can't collide with files git doesn't know about. The refusal message surfaces the porcelain summary so you can see exactly what to commit/stash.

Safety:

• git pull --ff-only so a non-fast-forward upstream aborts instead of producing a surprise merge.
• Refuses to pull when the working tree has modified TRACKED files unless --allow-dirty is passed.
• Any step's failure short-circuits the rest, so a half-applied update never reaches the restart phase.
• Source checkout and runtime data root (KESTREL_HOME) are resolved separately. The pull/install run against the source checkout (discovered by introspecting kestrel_sovereign.__file__ — must have both pyproject.toml and .git). The runtime data root is used only for feature sync's manifest lookup.

Pip-installed users (no editable source checkout — pip install kestrel-sovereign against PyPI) — both pull AND install are silently skipped; feature sync and restart still run. Upgrade the package itself with pip install --upgrade kestrel-sovereign first, then run kestrel update to pick up new features + restart.

Bootstrap when your installed kestrel update is older than the fix you want. If your CLI predates a kestrel update change you're trying to pull in, the old version may refuse on a dirty tree (early behaviour treated untracked files as dirty) or run the wrong install command. One-time bootstrap:

git pull && kestrel update          # explicit pull first, then the new code runs
# or
kestrel update --allow-dirty         # bypass the old refusal a single time

After the new code lands in your checkout, kestrel update alone handles subsequent refreshes.

Feature management (kestrel feature)

Kestrel ships a lean core. Optional feature packages register Feature classes through the kestrel_sovereign.features entry-point group. Provider packages, such as cloud, voice, and storage backends, register with provider-specific entry-point groups and are consumed by their owning core or feature module.

kestrel feature list                   # Show installed + available features
kestrel feature info <name>            # Detailed info about a feature
kestrel feature install <name>         # Install a feature package
kestrel feature upgrade [<name>...]    # Upgrade installed feature/provider packages
kestrel feature sync                   # Restore the host's declared feature set (see below)
kestrel feature enable <name>          # Enable an installed feature
kestrel feature disable <name>         # Disable without uninstalling
kestrel feature scaffold <name>        # Generate a new feature package skeleton

The canonical inventory of features lives in KESTREL_FEATURES.md; the runtime registry is in kestrel_sovereign/data/feature_registry.toml.

Keeping features installed across uv sync (kestrel feature sync)

Feature and provider packages are deliberately not declared in pyproject.toml (that would invert the open-core dependency direction), so uv sync — which makes the venv match declared dependencies exactly — prunes them. After a sync or any environment rebuild, an installed feature can silently vanish (e.g. voice endpoints start returning 404). kestrel feature upgrade cannot bring it back, because a pruned package leaves no entry-point to discover.

kestrel feature sync is the restore counterpart. It reads a host-local manifest, .kestrel-host-features.toml (gitignored; template at .kestrel-host-features.toml.example), and reinstalls whatever is missing:

kestrel feature sync --capture         # Snapshot currently-installed extensions into the manifest
kestrel feature sync                    # Reinstall the manifest's packages that are missing
kestrel feature sync --dry-run          # Preview without installing
kestrel feature sync --manifest PATH    # Use a non-default manifest location

A manifest entry names a package (by registry name or raw dist name) and may pin an editable checkout and extras:

[[feature]]
name = "voice"
extras = ["local"]                      # also pull the on-device pipeline (Piper / faster-whisper)

[[feature]]
name = "github"
editable = "/path/to/kestrel-feature-github"

This never touches pyproject.toml, so installing kestrel-sovereign itself still pulls zero feature packages. The typical recovery after a prune is kestrel feature sync followed by a host/agent restart so the restored entry-points load.

Per-Agent Configuration

Each agent can have a kestrel.toml config file in its directory:

# agent_data/myagent/kestrel.toml
[agent]
name = "MyAgent"
port = 8888
host = "0.0.0.0"
log_level = "INFO"

Create or edit config:

kestrel config ./agent_data/myagent --init           # Create config
kestrel config ./agent_data/myagent --set-port 8899  # Change port
kestrel config ./agent_data/myagent --set-name MyAgent  # Change name

Running Multiple Agents

Each agent runs on its own port. Create configs for each:

# Agent 1: Alpha on port 8888
kestrel create Alpha --port 8888
kestrel start Alpha

# Agent 2: Helper on port 8889
kestrel create Helper --port 8889
kestrel start Helper

# Check status of all agents
kestrel status

Alternative: Direct Commands

# Start server directly (set KESTREL_DB_PATH first).
# Pip-installed users invoke the in-package module (the wheel only ships
# `kestrel_sovereign/`, no top-level `server.py`):
KESTREL_DB_PATH=./agent_data/myagent uvicorn kestrel_sovereign.server:app --port 8888

# CLI chat (no server needed)
python -m kestrel_sovereign.main ./agent_data/myagent

# Source-clone shorthand: a thin re-export shim at the repo root keeps
# the historical commands working when you've cloned the repo:
#   uvicorn server:app --port 8888
#   python main.py ./agent_data/myagent

Note: KESTREL_DB_PATH is a directory path, not a file path. The database file kestrel_prime.db is created inside the specified directory. For example, setting KESTREL_DB_PATH=./agent_data/myagent stores the database at ./agent_data/myagent/kestrel_prime.db.

Keeping the host awake (long sessions / 24/7). Kestrel does not manage host power state itself — a long-running agent that must survive an idle machine relies on the OS staying awake. For a dedicated deployment, prefer a service supervisor (launchd/systemd) or an always-on host (Mac mini, VPS, cloud container). For a foreground run on a laptop that would otherwise sleep, wrap the command in the platform's keep-awake utility:

• macOS: caffeinate -s <command> (e.g. caffeinate -s kestrel start MyAgent). -s prevents idle sleep while on AC power.
• Linux (systemd): systemd-inhibit --what=sleep:idle --why="kestrel agent" <command> holds a sleep inhibitor for the command's lifetime.
• Windows: there is no built-in command wrapper. Either disable sleep under Settings → System → Power, or hold an execution-state assertion for the session, e.g. in PowerShell:

  $sig = '[DllImport("kernel32.dll")] public static extern uint SetThreadExecutionState(uint e);'
  $t = Add-Type -MemberDefinition $sig -Name Power -Namespace Win32 -PassThru
  $t::SetThreadExecutionState(0x80000003)  # ES_CONTINUOUS | ES_SYSTEM_REQUIRED
  kestrel start MyAgent
  

🖥️ Web UI (Sovereign Console)

Kestrel includes a built-in web interface called the Sovereign Console. Once your agent is running, open the URL the CLI printed on start — http://localhost:8888 for the multi-agent host (default kestrel start mode), or the per-agent port for a single-agent start — in any browser; no additional software required.

The console provides 9 tabs:

Tab	Description
Chat	Converse with the agent (supports model selection, privacy modes, chat history)
Identity	View the agent's DID, name, and cryptographic identity
Constitution	View and audit the agent's constitutional principles
Memories	Browse the agent's knowledge graph and stored memories
Tasks	Monitor background tasks and activity
Sovereignty	Manage data sovereignty, backups, and exports
Resources	View agent resource usage and configuration
Features	Browse installed, available, enabled, and disabled features
Security	Manage permissions, audit logs, and session security

Alternative clients: The server also exposes an OpenAI-compatible API at /v1/chat/completions, so you can connect any OpenAI-compatible client (e.g., Open WebUI) if you prefer.

🏗️ Architecture Overview

Kestrel agents are built on several key components:

• Cryptographic Identity: Each agent has a unique DID (Decentralized Identifier)
• Enhanced Storage: Local-first memory with SQL-backed stores, SQLAlchemy vector mappings, FTS, knowledge graphs, and RAG; provider-standard embedding generation is in progress
• Multi-Model LLM: Fallback between local (Ollama) and cloud (OpenAI) models
• Constitutional Governance: Immutable principles with interpretive flexibility
• Blockchain Anchoring: Optional integrity verification via blockchain

📁 Project Structure

kestrel-sovereign/
├── kestrel_sovereign/         # Core sovereign package
│   ├── cli.py                 # `kestrel` CLI entry point (canonical)
│   ├── kestrel_agent.py       # Core agent class
│   ├── inception_service.py   # Agent creation (DID + genesis audit)
│   ├── agent_config.py        # Per-agent config loader
│   ├── data/feature_registry.toml  # Runtime feature registry
│   ├── features/              # Core bundled features
│   ├── storage/               # SQL-backed storage facade and stores
│   ├── static/                # Sovereign Console frontend
│   └── ...
├── server.py                  # Re-export shim → kestrel_sovereign/server.py
├── host.py                    # Multi-agent multi_agent host
├── main.py                    # Re-export shim → kestrel_sovereign/main.py
├── docs/                      # Architecture & guides
└── tests/                     # Test suite

# The Kestrel SDK lives in its own repo + PyPI package:
#   kestrel-sovereign-sdk  (https://github.com/KestrelSovereignAI/kestrel-sovereign-sdk)
# Feature authors import its types directly:
#   from kestrel_sdk.features.base import Feature, tool
#   from kestrel_sdk.tools.base   import ToolCategory, ToolResult
#   from kestrel_sdk.hooks.base   import Hook, HookEvent

🎯 Core Features

1. Sovereign Memory

• Persistent Storage: SQL-backed stores with full-text search and knowledge graphs
• RAG Pipeline: Document chunking and semantic retrieval through SQLAlchemy-backed vector search where available; embedding generation still uses the current embedding service while provider-standard embedding functions are being added
• Conversation History: Complete interaction tracking with metadata
• Human-Led Interactions: Prioritizes user narratives (e.g., storytelling) for preservation and no-loss continuity.

2. Multi-Model Intelligence

• Local First: Ollama for privacy and cost efficiency
• Cloud Fallback: OpenAI for complex reasoning when needed
• Configurable: Easy provider switching via configuration

3. Cryptographic Identity

• DID Generation: Unique decentralized identifiers
• Signed Operations: Cryptographic verification of agent actions
• Ownership Transfer: Secure agent handoff between users

4. Constitutional Governance

• Immutable Articles: Core principles that cannot be changed
• Interpretive Canons: Flexible guidelines for decision-making
• Amendment Process: Cryptographically-signed governance updates

5. Data Sovereignty & Privacy Modes

• Ephemeral Mode: True off-the-record conversations (nothing stored)
• Privacy Granularity: 5 distinct privacy levels for different use cases
• Decentralized Storage: Filecoin/IPFS integration for vendor independence
• Optional Economics: Wallet and autonomous payment flows are installable feature-package surfaces, not part of the base install

⚠️ Feature Stability (v0.18+ Beta)

Kestrel covers a wide surface; not all of it ships at the same maturity. Verified 2026-05-31 against the current feature inventory, runtime registry, package-boundary docs, tests, and recent documentation audit:

✅ Stable — battle-tested in production by the maintainers

• Constitutional AI — Genesis audits, hierarchical permissions, approval queues
• DID-based Identity — did:pkh format, portable agent identity, export/import
• 5-Level Privacy Modes — EPHEMERAL → ISOLATED → ANONYMOUS → NORMAL → PUBLIC
• Memory & Storage — Local-first memory, FTS, knowledge graph, RAG, and SQLAlchemy-backed vector storage are core; embedding generation is the moving part as LLM providers gain standardized embedding functions
• LLM service — Vendor/route/model architecture with Anthropic, OpenAI, Vertex AI, Ollama, OpenRouter, xAI, Groq; retry, structured output, streaming, vision
• A2A Protocol — JSON-RPC 2.0 for agent-to-agent communication
• Cloud Run deploy — 90 tests, active maintenance; the most-tested cloud feature

🧪 Experimental — works on the happy path; gaps to know about

• Optional voice feature package — kestrel-feature-voice supplies VoiceFeature; cloud TTS/STT providers live in kestrel-voice-* provider packages.
• Wallet / agent economics — installable feature package surface, not part of the base install.
• RunPod, Vast.ai, and GCP Compute — cloud provider packages/bridges, not core features; integration tests skip in CI without provider credentials.
• Azure Container Apps deploy — provider stub; not the recommended deploy target.
• GitHub code introspection — file reading, code search, definition lookup, issue tools all work (48 unit tests). The deeper static-analysis surface promised in docs/architecture/GITHUB_FEATURE_DESIGN.md (call graphs, inheritance trees, dependency analysis) is not implemented.
• Training (LoRA pipeline) — core ships the protocol + factory; the local-MPS adapter is actively maintained. Cloud-training adapters (RunPod/Vertex/Replicate) work but skip CI without API keys; production-grade adapters are being moved to private packages.

⚠️ Work-in-progress

• DID Verification Layer — generation works; verification is incomplete
• E2E Test Stability — some integration tests are occasionally flaky
• API Stability — APIs may change before v1.0; breaking changes will be documented

❌ Not implemented in this framework

These are not bundled in the kestrel-sovereign base install:

• Turnkey channel adapters — WhatsApp, Telegram, Discord, and Slack adapters are not shipped in the base install; the core channels surface is a registry/logging foundation.
• Bundled voice cloud backends — ElevenLabs, Deepgram, and OpenAI voice support live in optional kestrel-voice-* provider packages.
• Browser Automation — Chrome/Chromium control
• Visual Workspaces — A2UI canvas, live reload

Bottom line: Kestrel is ready for developers building privacy-first, economically-independent AI agents and for the soft-launch preview cohort. Not yet ready for unmanaged production apps or general consumer use. If you find a stability classification above doesn't match your experience, please open an issue — that's the kind of signal we need.

📚 Documentation

Detailed documentation is available in the docs/ directory:

• Documentation Index

• Agent Ecosystem

• Agent Economics

• Kestrel Constitution

• Cryptographic Anchoring

• Decentralized Storage

• Multi-Model Support

• Privacy Modes

• LLM Service Architecture

💡 Example Applications

Kestrel is a foundation for AI agents that need to outlive any single vendor, deployment, or owner. Concrete deployments and good-fit use cases:

• Healthcare RPM agents — Constitutional governance over an LLM, persistent patient-owned memory, audit trail for every clinically-relevant action.
• Long-running personal research agents — Memory accumulates across months without dependency on a single provider's chat history.
• Custodial agents for sensitive document workflows — Privacy-mode tiers (EPHEMERAL → PUBLIC) let one agent handle both an off-the-record consult and a fully-anchored long-term contract.
• Multi-agent A2A networks — JSON-RPC 2.0 agent-to-agent protocol lets sovereign agents collaborate without surrendering their identity to a central broker.

🧪 Testing

Run the test suite from the activated virtual environment:

# Run a single test with uv and -x
uv run pytest -x tests/test_inception.py::test_successful_inception

Clean Install Verification

Kestrel supports multiple installation configurations. Use the verification script to test that clean installs work correctly across all supported scenarios:

# Run all 5 install scenarios (creates isolated venvs)
uv run kestrel verify-install

# Run specific tests only
uv run kestrel verify-install 1 3    # SDK-only and wallet package

The install matrix covers:

Test	Scenario	Verifies
1	SDK only	from kestrel_sdk.features.base import Feature
2	Core sovereign	from kestrel_sovereign.features.base import Feature compatibility + /health
3	Feature package	from kestrel_feature_wallet import WalletFeature
4	SDK + feature dev mode	Feature packages can develop against SDK alone
5	Full stack	Sovereign + wallet + intelligence, entry_point discovery

Integration tests for the same import paths run as part of the normal test suite:

uv run pytest tests/integration/test_clean_install_verification.py -v

🔧 Configuration

LLM Configuration (kestrel.toml [llm])

LLM config lives under the [llm] section of kestrel.toml. The setup wizard (kestrel setup llm) will write it for you; you can also hand-edit kestrel.toml after copying from kestrel.toml.example.

Kestrel uses a vendor/route/model schema. A vendor is who makes the weights; a route is how to reach them (adapter + base URL + auth). API keys belong in .env and are referenced by api_key_env. See kestrel.toml.example and docs/architecture/LLM_SERVICE_ARCHITECTURE.md for the canonical spec.

[llm]
route_priority = ["openai:api", "ollama:local"]

[llm.vendors.openai]
is_cloud = true

[llm.vendors.openai.routes.api]
adapter        = "OpenAIAdapter"
api_key_env    = "OPENAI_API_KEY"
model          = "auto"
selection_hints = ["gpt-5", "mini"]

[llm.vendors.ollama]
is_cloud = false

[llm.vendors.ollama.routes.local]
adapter        = "OllamaAdapter"
host           = "http://localhost:11434"
model          = "auto"
selection_hints = ["llama3.2", "qwen"]

Pre-2026-05 setups used a standalone llm_config.toml at the repo root. That path was removed (epic #938). Run kestrel migrate-llm-config to fold a legacy file into kestrel.toml [llm]; the source is renamed to .bak, your prior kestrel.toml is timestamp-backed-up, and the operation is idempotent.

Environment Variables

See .env.example for a complete list. Key variables:

LLM Providers:

• OPENROUTER_API_KEY: OpenRouter API key (recommended - access to multiple providers)
• OPENAI_API_KEY: OpenAI API key for cloud models
• ANTHROPIC_API_KEY: Anthropic API key for Claude models

Storage:

• KESTREL_DB_PATH: Directory where the agent database is stored (default: ./agent_data). This is a directory path -- the database file kestrel_prime.db is created inside it.
• KESTREL_DATA_KEY: Fernet encryption key for data at rest

GitHub Integration:

• GITHUB_TOKEN: Personal access token for GitHub features
• GITHUB_SELF_REPO: Agent's source repository (default: KestrelSovereignAI/kestrel-sovereign)

🚢 Deployment

Kestrel supports multiple deployment targets. See KESTREL_FEATURES.md for the full catalog.

Cloud Run (Serverless)

Scales to zero when idle ($0/month), auto-scales under load. Each sovereign agent gets its own service.

# One-time: set up GCP secrets from .env
uv run kestrel deploy secrets sync

# Build and push to GCR (both single-agent and multi_agent images)
uv run kestrel deploy build

# Deploy to dev (multi-agent host, always warm) or prod (single-agent, auto-scales)
uv run kestrel deploy dev
uv run kestrel deploy prod

Profiles live in deploy_config.toml. See docs/deployment/README.md for the full runbook (status, logs, teardown, health).

Auto-deploys on version tags via GitHub Actions.

Docker (Local)

# Remote LLM — smallest image (~500MB)
docker build -f docker/Dockerfile.remote -t kestrel .
docker run -p 8888:8888 -e OPENAI_API_KEY=... kestrel

# Standalone with Ollama (no API keys needed)
docker build -f docker/Dockerfile.standalone -t kestrel-standalone .
docker run -p 8888:8888 kestrel-standalone

# GPU with CUDA
docker build -f docker/Dockerfile.gpu -t kestrel-gpu .
docker run --gpus all -p 8888:8888 kestrel-gpu

🔐 Backups and Storage Tiers

Backups can be created interactively from the agent using privacy-gated storage tiers:

• local: cache the backup tar.gz locally only
• ipfs: encrypt + gzip and store on IPFS; also cache locally
• filecoin: same as IPFS and propose a Filecoin deal via Lotus when available; fallback to local if not

Privacy gating:

• EPHEMERAL: backups disabled
• ISOLATED: cache-only; use !promote-backup to save the isolated session and back up
• ANONYMOUS: backups allowed; encryption forced for filecoin tier
• NORMAL: backups allowed; encryption configurable (default on)

Usage from the REPL:

!backup tier=local
!backup tier=ipfs
!backup tier=filecoin
!promote-backup tier=filecoin

Each backup produces a backup_artifact node in the graph linked to the agent with properties like content_hash, ipfs_cid, filecoin_deal_id, encrypted, and timestamp.

🔒 Encryption at Rest

• Files and conversation history can be encrypted at rest by setting KESTREL_DATA_KEY (Fernet key or passphrase):

export KESTREL_DATA_KEY=$(python - <<'PY'
from cryptography.fernet import Fernet
print(Fernet.generate_key().decode())
PY
)

• With the key set, stored file blobs and conversation entries are encrypted transparently. Backups remain encrypted by default. For production, wire the backup master key to an env/KMS and avoid the dev placeholder.

Optional: Full-DB Encryption (SQLCipher)

• If you install pysqlcipher3 and set KESTREL_DB_KEY, the SQLite connection will use SQLCipher and encrypt the entire DB:

export KESTREL_DB_KEY="your-db-passphrase"
uv run python -m kestrel_sovereign.server

• Without pysqlcipher3, the system falls back to normal SQLite. File blobs and conversations still encrypt with KESTREL_DATA_KEY if set.

🧩 OpenAI-Compatible API

The server exposes OpenAI-compatible endpoints for use with third-party clients:

• GET /v1/models
• POST /v1/chat/completions

For most users, the built-in Sovereign Console at the printed URL (http://localhost:8888 for the multi-agent host, or the agent's own port for a single-agent start) is the easiest way to interact with your agent (see the Web UI section above). If you prefer an external client, point any OpenAI-compatible tool (e.g., Open WebUI) at your server's /v1/chat/completions endpoint. Use the model name from /v1/models.

Auth: every request to /v1/... requires the X-API-Key header (or Authorization: Bearer <key>). The key was written to .env as KESTREL_API_KEY by kestrel setup (or --quickstart). Most OpenAI-compatible clients let you set the key via their OPENAI_API_KEY env var or settings UI; point that at your KESTREL_API_KEY value.

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Add tests for new functionality
4. Run the test suite: python -m pytest -x
5. Submit a pull request

📄 License

Apache 2.0 — see LICENSE for details.

🆘 Support

• Issues: GitHub Issues for bug reports and feature requests
• Discussions: GitHub Discussions for questions and ideas
• Documentation: See docs/ and KESTREL_FEATURES.md

---

Kestrel: Where AI meets sovereignty.

📚 Key Files Reference

File	Purpose
kestrel_sovereign/cli.py	Canonical kestrel CLI entry point
kestrel_sovereign/server.py	FastAPI agent server (root server.py is a re-export shim for source clones)
host.py	Multi-agent multi_agent host (Cloud Run)
kestrel_sovereign/main.py	Direct interactive REPL (root main.py is a re-export shim for source clones)
kestrel.toml	Unified config (LLM, agents, features). [llm] holds provider config.
KESTREL_FEATURES.md	Canonical feature inventory
kestrel_sovereign/kestrel_agent.py	Core agent logic
kestrel_sovereign/agent_config.py	Per-agent config loader
kestrel_sovereign/inception_service.py	New agent creation (DID + genesis audit)
kestrel_sovereign/data/feature_registry.toml	Runtime feature registry
agent_data/<name>/kestrel.toml	Per-agent configuration
agent_data/<name>/kestrel_prime.db	Agent database
docs/**/*.md	Detailed documentation

Architecture

Storage System

The Kestrel storage system is designed to be modular and extensible. It is composed of several specialized components, orchestrated by a high-level facade.

• storage.Database / AsyncDatabase: Manages the low-level SQL-backed connection and schema. SQLite remains the local default; PostgreSQL is supported through the async backend layer.
• storage.FileStore: Handles the storage and retrieval of files.
• storage.GraphStore: Manages the knowledge graph (nodes and edges).
• storage.RAGStore: Responsible for document chunking and semantic search for the RAG pipeline and "case law" system. It uses SQLAlchemy/vector backends when available and falls back gracefully; embedding generation is being aligned with standardized provider embedding functions.
• storage.ConversationStore: Manages the agent's conversation history.

The main Storage class in storage/__init__.py acts as a facade, providing a single, unified interface to these components.

Genesis Self-Audit

To ensure the integrity of all new agents, Kestrel implements a "genesis self-audit." When a new agent is created via inception_service.py:

1. The agent's foundational files (keys, database) are created.
2. The KESTREL_CONSTITUTION.md is stored as the agent's first memory.
3. The agent is instantiated and its very first action is to perform an integrity audit on its own constitution.
4. If the audit returns a high risk level, the creation process is aborted, and all generated files are cleaned up, preventing the existence of a non-compliant agent.

This process guarantees that every agent in the ecosystem starts from a foundation of verifiable integrity.

🔄 Next Steps

After getting started:

1. Explore Features: Read KESTREL_FEATURES.md and docs/guides/BUILDING_FEATURES.md