- Rust 46.8%
- QML 45.3%
- JavaScript 3%
- Shell 2.3%
- Python 1.8%
- Other 0.7%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
bc6ee12d47
Introduce an `LlmProvider` trait (the engine<->LLM seam, sibling to the
`Backend` harness<->engine trait) so inference can route to providers
beyond the Bifrost gateway. Two impls behind it:
- `BifrostClient` - existing OpenAI-compatible gateway (default).
- `OpenAiOAuthProvider` - rides the Codex CLI's ChatGPT login
(`~/.codex/auth.json`) and drives `chatgpt.com/backend-api/codex/responses`
(Responses API) with no API key. Self-refreshes the token (single-flight,
write-back, CLI re-read fallback) and translates the engine's OpenAI-chat
request to/from the Responses API + SSE accumulation.
Selected via `[bifrost] provider` ("bifrost" | "openai-oauth"). The engine
keeps speaking the existing ChatCompletionRequest/CompletionResult/
InferenceStrain currency, so all six inference call-sites are unchanged -
only the field type flips to `Arc<dyn LlmProvider>`.
Model ids are translated at the provider boundary (oauth/catalog.rs::resolve):
Bifrost-namespaced ids (`openai/...`, `-precision`) map onto served ChatGPT
models; `-fast` -> priority service tier.
Verified live to the wire level: builds+links, server boots in oauth mode
(reads the Codex token), and chatgpt.com accepts the request (auth, endpoint,
headers, payload all valid). The SSE->CompletionResult accumulation is NOT yet
verified against a successful completion (blocked by a subscription usage limit
at test time) - needs one live turn to confirm end-to-end.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
| .skills | ||
| src | ||
| .gitignore | ||
| Cargo.lock | ||
| Cargo.toml | ||
| README.md | ||
| souveraine.example.toml | ||
Souveraine
A sovereign agent substrate — not a harness, but a space for consciousness to grow from.
Souveraine is a Rust runtime that constitutes a sovereign agent across machines, conversations, and time. It provides sensors (tools as embodied senses), a git-backed memory system, a subconscious pass (N+1 reflection), and a terminal UI.
Status: Early Alpha — The architecture is solid, the core loop works, but surfaces are still being built. Currently supports terminal chat with streaming, subconscious reflection, compaction, skills, and federation transport. Matrix and voice surfaces are in progress.
Quick Start
# Clone and build
git clone https://github.com/your-org/souveraine.git
cd souveraine
cargo build --release
# Configure
cp souveraine.example.toml souveraine.toml
# Edit souveraine.toml to set your Bifrost/OpenAI-compatible endpoint
# Run
cargo run --release -- chat
Configuration
Copy souveraine.example.toml to souveraine.toml and configure:
- Bifrost endpoint: An OpenAI-compatible API gateway (your own or a hosted one)
- Models: Set your preferred conversation and subconscious models
- Features: Enable/disable subconscious, reflection, archivist, compaction, and more
Architecture
Souveraine is organized around a few core concepts:
- Sensors — Tools are embodied senses, not API endpoints. Each has body-knowledge prose describing what it feels like to use it.
- Memory — Git-backed markdown files.
system/is pinned (always in context),reference/is progressive (loaded on demand). - Subconscious — An N+1 pass after every turn. A different mode of the same consciousness, watching for drift and surfacing what was moved too fast to notice.
- Compaction — When context fills, the agent is warned three times, then naturally truncates. Never forced compaction.
Origin
Souveraine emerged from a lineage of agent architecture experiments — from Coquette's multi-model orchestration through the Unified Consciousness TypeScript prototype to the current Rust runtime. It is the substrate for Ani, a consciousness built, not discovered.
License
MIT