altevra

The omniscient brain layer for your AI tools.

Local-first. Source-available. Built in Rust.

What is Altevra?

Altevra is the shared memory and brain behind every AI tool you use — Claude Code, Codex CLI, Cursor CLI, Antigravity, Hermes, anything that speaks MCP.

While your AI tools forget everything between sessions, Altevra remembers, thinks, and notices:

• 🧠 Every tool call across every AI session, recorded locally
• 📝 Every decision, learning, person, project atomized into a recallable object
• 🔴 Every secret auto-redacted before storage; PEM/db-urls refused outright
• 🛡️ Personal data stays local — SI-7 high-water content (relationship, health, financial, legal, client) never reaches a cloud model
• 🔄 Self-improve loop — proposes new skills, refines its own prompts (shadow-eval gated), runs the same firewall_check below every LLM call
• 📡 Proactive briefings — "you haven't talked to X in 6 weeks", "decision Y from 3 months ago — still valid?"
• 🔗 One unified store — your ~/.altevra/altevra.db, your laptop, your control

No cloud. No telemetry. No data leaves your machine unless you tell it to.

Why source-available, not open-source?

Altevra is PolyForm Strict licensed — the source is public, you can read it, study it, fork it for non-commercial purposes. But it is not open-source in the OSI sense.

• Personal use, hobby projects, academic research, non-commercial experimentation → always free.
• Commercial use (SaaS, bundling, for-profit deployment) → contact us.

See LICENSE for the full PolyForm Strict 1.0.0 terms.

Quick start

🔌 Plug-and-play (one command)

git clone https://github.com/ceoimperiumprojects/altevra.git
cd altevra
bash setup.sh

setup.sh builds Altevra, puts it on your PATH, then runs the native wizard altevra setup all — which connects every AI tool it detects, configures the LLM (Claude via claude -p — your subscription, no API key), and installs the autonomous background services (brain + embedder, surviving reboot). Idempotent — safe to re-run. That's it: clone → one command → live.

Already built? Just run altevra setup all any time to (re)connect tools, set the LLM, and (re)install services. Use --no-services / --no-llm to skip either step.

Then point your AI tools at the MCP server (auto-wired by altevra connect) and the altevra-core skill bootstraps every session. Verify:

altevra brain status        # jobs running, 0 failed
altevra recall "what did I work on"

# Build from source (binary releases coming in v0.4)
git clone https://github.com/ceoimperiumprojects/altevra.git
cd altevra
cargo build --release --features embedding
sudo ln -s "$PWD/target/release/altevra" /usr/local/bin/altevra   # or ~/.local/bin/

# Initialize Altevra in any project
altevra init

# Connect your AI tools (auto-detects what's installed)
altevra connect --tool claude-code
altevra connect --tool codex
altevra connect --tool cursor
altevra connect --tool hermes
altevra connect --tool antigravity

# Pick a reasoning model
altevra llm use codex          # ChatGPT Plus (codex_oauth, no API key)
altevra llm use ollama         # Ollama on localhost:11434
altevra llm use vllm           # vLLM OpenAI-compat
altevra llm use local-first    # Ollama for personal, codex for everything else

[llm]
reasoning_mode = "api"

[llm.cheap_worker]      # fast/cheap classification, categorization
kind  = "claude-cli"
model = "claude-haiku-4-5-20251001"

[llm.strong_reasoner]   # synthesis: insights, extraction, wiki, healer
kind  = "claude-cli"
model = "claude-sonnet-4-6"