docs(release): v0.0.14 sync — Gemini priority + 10 new features documented

ARCHITECTURE.md

@@ -49,7 +49,7 @@ If it's app-layer complexity → reject it. If it strengthens sensor drivers →
 │  │  Episodic (NDJSON/FTS5) | Semantic (SQLite)         │  │
 │  └──────────────────────┬─────────────────────────────┘  │
 │  ┌──────────────────────┴─────────────────────────────┐  │
-│  │  MCP Server (42 tools) — sole control interface     │  │
+│  │  MCP Server (47 tools) — sole control interface     │  │
 │  │  RBAC (Viewer/Operator/Admin) | Bearer token auth   │  │
 │  │  PlatformOperations trait | Rate limiting            │  │
 │  └──────────────────────┬─────────────────────────────┘  │
@@ -332,7 +332,7 @@ BUBBALOOP_ZENOH_ENDPOINT=tcp/127.0.0.1:7447  # Optional override
 
 ## MCP Server
 
-MCP is the **sole control interface**. 42 MCP tools + agent-internal tools across categories:
+MCP is the **sole control interface**. 47 MCP tools + ~43 agent-internal tools across categories:
 
 | Category | Tools |
 |----------|-------|
@@ -347,7 +347,7 @@ MCP is the **sole control interface**. 42 MCP tools + agent-internal tools acros
 | **Missions** | list_missions, pause_mission, resume_mission, cancel_mission — YAML-file-driven (`~/.bubbaloop/agents/{id}/missions/`) |
 | **Constraints** | register_constraint, list_constraints — per-mission safety limits; params_json formats: `workspace={"x":[-1,1],"y":[-1,1],"z":[0,2]}`, `max_velocity=1.5`, `forbidden_zone={"center":[0,0,0],"radius":0.3}`, `max_force=50.0` |
 | **Alerts** | register_alert, unregister_alert — reactive arousal triggers when world state predicate matches |
-| **Agent-internal** | read_file, write_file, run_command, memory_search, memory_forget, schedule_task, create_proposal, publish_to_topic, get_system_telemetry, get_telemetry_history, update_telemetry_config |
+| **Agent-internal** | grab_frame (Zenoh-query camera node for one JPEG → fed to vision-capable provider), read_file, write_file, run_command, memory_search, memory_forget, schedule_task, create_proposal, publish_to_topic, get_system_telemetry, get_telemetry_history, update_telemetry_config |
 
 ### Transport Options
 
@@ -435,12 +435,12 @@ MCP is the **sole control interface**. 42 MCP tools + agent-internal tools acros
 | Runtime | Rust + Tokio | Memory safety, small binary, edge-ready |
 | Data plane | Zenoh | Zero-copy pub/sub, decentralized, Rust-native |
 | Schemas | Protobuf + prost | Self-describing, runtime introspection |
-| Control | MCP (rmcp) | Standard AI agent interface, 42 MCP tools + agent-internal tools |
+| Control | MCP (rmcp) | Standard AI agent interface, 47 MCP tools + ~43 agent-internal tools |
 | Memory | SQLite (rusqlite) + NDJSON | 4-tier: world state (live SQLite) + RAM + episodic (NDJSON/FTS5) + semantic (SQLite). World state updated by context providers, not LLM. |
 | CLI | argh | Minimal, fast compile |
 | Logging | log + env_logger | Simple, stderr-only |
 | systemd | zbus (D-Bus) | No subprocess spawning, safe |
-| LLM | ModelProvider trait (reqwest) | Claude (OAuth + API key) and Ollama (local, tool calling) |
+| LLM | ModelProvider trait (reqwest) | Gemini (recommended — free tier, vision), Claude (OAuth + API key, vision), Ollama (local, tool calling, text-only) |
 | HTTP | axum | Dashboard + MCP HTTP transport |
 | Telemetry | sysinfo | CPU/RAM/disk monitoring, cross-platform |
 

CHANGELOG.md

@@ -1,5 +1,87 @@
 # Changelog
 
+## [0.0.14](https://github.com/kornia/bubbaloop/compare/v0.0.13...v0.0.14) (2026-05-19)
+
+### Features
+
+* **agent:** `grab_frame` tool — agent fetches a single JPEG frame from a camera node and feeds it to a vision-capable provider for analysis. Multi-camera fan-out supported. ([ff7fe08](https://github.com/kornia/bubbaloop/commit/ff7fe08), [85dff9c](https://github.com/kornia/bubbaloop/commit/85dff9c))
+* **agent:** Gemini Vision provider (`crates/bubbaloop/src/agent/provider/gemini.rs`) — recommended default for new users (free tier, strong vision support). ([ff7fe08](https://github.com/kornia/bubbaloop/commit/ff7fe08))
+* **agent:** client-side turn cancellation via Zenoh `bubbaloop/global/{machine_id}/agent/{agent_id}/cancel` topic; Ctrl-C in the REPL aborts the in-flight turn. ([2ea9176](https://github.com/kornia/bubbaloop/commit/2ea9176), [ac384dc](https://github.com/kornia/bubbaloop/commit/ac384dc))
+* **dashboard:** Chat tab — talk to agents from the browser via HTTP+SSE to the daemon's agent gateway. Includes tool-call trace rendering and image-in-chat from `grab_frame`. ([4140a1a](https://github.com/kornia/bubbaloop/commit/4140a1a))
+* **chat:** multi-provider login status (Gemini / Claude / Ollama), 404 on unknown agent, Claude OAuth risk warning. ([fe76d67](https://github.com/kornia/bubbaloop/commit/fe76d67))
+* **mcp:** `bubbaloop mcp --token` prints the stored bearer token so `.mcp.json` can pick it up without a race. ([4dadb64](https://github.com/kornia/bubbaloop/commit/4dadb64))
+
+### Bug Fixes
+
+* **agent:** `grab_frame` reads binary JPEG payload + JSON metadata attachment from camera nodes (was: base64-in-JSON). ([326a9dd](https://github.com/kornia/bubbaloop/commit/326a9dd))
+* **chat:** clear Responding state after multi-camera `grab_frame` completes. ([85dff9c](https://github.com/kornia/bubbaloop/commit/85dff9c))
+* **mcp:** eliminate token race between gateway and agent runtime. ([d03efba](https://github.com/kornia/bubbaloop/commit/d03efba))
+
+### Chores
+
+* **dashboard:** remove LibraryView component (replaced by Chat tab). ([5d687be](https://github.com/kornia/bubbaloop/commit/5d687be))
+
+### Docs
+
+* Prioritize Gemini as the recommended default provider across README, agent-guide, quickstart, configuration, and CLI reference (#100).
+* Sync CHANGELOG, CLAUDE.md, ARCHITECTURE.md, ROADMAP.md, concept docs with v0.0.14 features.
+
+## [0.0.13](https://github.com/kornia/bubbaloop/compare/v0.0.12...v0.0.13) (2026-05-04)
+
+### Features
+
+* **mcp:** `install_node` accepts name + config flags (#87)
+
+Released as `chore(release): v0.0.13` (#88) — install-flow polish from the e2e exercise.
+
+## [0.0.12](https://github.com/kornia/bubbaloop/compare/v0.0.11...v0.0.12) (2026-05-04)
+
+### Features
+
+* **agent:** recorder pipeline overhaul — MCP plugin + production refinements + install + SDK testing (#85)
+* **agent:** Phase 3 — reactive alerts end-to-end (rule-based world-state monitoring with debounce, circuit breaker, world-state sweeper, hard reqwest timeout for stuck LLM generations)
+* **sdk:** `publisher_cbor(local=True)` for CBOR+SHM in the Python SDK (#78)
+
+### Refactors
+
+* **wire-format:** remove protobuf in favor of CBOR + manifest schemas across the dataplane (#76)
+
+## [0.0.11](https://github.com/kornia/bubbaloop/compare/v0.0.10...v0.0.11) (2026-04-06)
+
+### Features
+
+* **agent:** Physical AI Memory & Mission system — scheduled tasks for recurring actions, missions vs tasks distinction
+* **tui:** real-time agent context visibility with System event type; show tool results by default (full detail with `-v`)
+
+### Bug Fixes
+
+* **agent:** prevent 400 'unexpected tool_use' errors
+* **tui:** drain subscriber and improve tool-result contrast
+* **build:** binary size reduction (`panic=abort`, `rusqlite` bundled, axum without `ws`)
+
+## [0.0.10](https://github.com/kornia/bubbaloop/compare/v0.0.9...v0.0.10) (2026-03-30)
+
+### Features
+
+* **agent:** `bubbaloop agent setup` command with first-run onboarding (#51)
+
+## [0.0.9](https://github.com/kornia/bubbaloop/compare/v0.0.8...v0.0.9) (2026-03-29)
+
+### Features
+
+* **agent:** OpenClaw-inspired agent rewrite with multi-agent runtime (#47)
+* **agent:** minimal opt-in Ollama support for local inference (#49)
+
+### Docs
+
+* Premium theme redesign, factual fixes against Rust source, agent-first narrative
+
+## [0.0.8](https://github.com/kornia/bubbaloop/compare/v0.0.7...v0.0.8) (2026-03-10)
+
+### Features
+
+* **cli:** Claude subscription login via `setup-token` (#36)
+
 ## [0.0.7](https://github.com/kornia/bubbaloop/compare/v0.0.6...v0.0.7) (2026-03-01)
 
 ### Features

CLAUDE.md

@@ -28,12 +28,15 @@ Key source files in `crates/bubbaloop/src/`:
 - `agent/runtime.rs` — Multi-agent runtime: AgentsConfig, AgentRuntime, ZenohSink, agent_loop; `agent_directory()` (pub)
 - `agent/prompt.rs` — System prompt builder; `build_system_prompt_with_soul_path()` injects onboarding prompt for new agents
 - `agent/soul.rs` — Soul struct, first-run onboarding, notify hot-reload (`~/.bubbaloop/soul/`)
-- `agent/provider/mod.rs` — ModelProvider trait, Message, ContentBlock, ToolDefinition, StreamEvent
+- `agent/provider/mod.rs` — ModelProvider trait, Message, ContentBlock (incl. `Image`), ToolDefinition, StreamEvent
+- `agent/provider/gemini.rs` — Gemini API client (recommended default — free tier, vision-capable)
 - `agent/provider/claude.rs` — Claude API client with dual auth (API key + OAuth bearer token)
 - `agent/provider/ollama.rs` — Ollama local LLM client with tool calling (`/api/chat`)
+- `agent/camera.rs` — `grab_frame` implementation: Zenoh query for one JPEG frame + JSON metadata attachment, emits `ContentBlock::Image`
+- `agent/gateway.rs` — Zenoh + HTTP+SSE agent gateway; per-agent cancel topic `bubbaloop/global/{machine_id}/agent/{agent_id}/cancel`
 - `agent/memory/` — 4-tier: world state (live SQLite) + short-term (RAM) + episodic (NDJSON) + semantic (SQLite)
 - `agent/heartbeat.rs` — Adaptive heartbeat: arousal + decay + state collection
-- `agent/dispatch.rs` — Internal MCP tool dispatch (42 MCP + agent-internal tools, includes telemetry + beliefs + constraints + missions)
+- `agent/dispatch.rs` — Internal tool dispatch (47 public MCP tools + ~43 agent-internal tools incl. `grab_frame`, telemetry, beliefs, constraints, missions)
 - `cli/node/mod.rs` — node CRUD, validation, list/add/remove
 - `cli/node/install.rs` — install, precompiled binary download, GitHub clone
 - `cli/node/lifecycle.rs` — start, stop, restart, logs

CONTRIBUTING.md

@@ -41,7 +41,7 @@ If a workflow only improves the workflow itself, **delete it**. The sensor nodes
 
 1. Check argh conventions in CLAUDE.md (`#[derive(FromArgs)]`, `#[argh(subcommand)]`)
 2. Write command + unit tests (co-located `#[cfg(test)] mod tests`)
-3. Verify 298+ Rust tests pass
+3. Verify 2400+ Rust tests pass
 4. Zero clippy warnings enforced
 
 **Security**: Node names `[a-zA-Z0-9_-]{1,64}`, no null bytes.
@@ -52,7 +52,7 @@ If a workflow only improves the workflow itself, **delete it**. The sensor nodes
 2. Update proto + rebuild BOTH descriptor pipelines (`bubbaloop-schemas` AND `bubbaloop`)
 3. Update MCP tool handlers in `mcp/mod.rs`, add integration tests
 4. Update dashboard types if applicable
-5. Full system check (298+ Rust + 47 MCP integration tests)
+5. Full system check (2400+ Rust unit tests + 47 MCP integration tests)
 
 **Critical**: Proto changes require rebuilding both descriptor pipelines.
 
@@ -80,7 +80,7 @@ Start at the lowest tier. Escalate only on failure. Haiku costs ~10x less than O
 | Command | When | Why |
 |---------|------|-----|
 | `pixi run check` | After every Rust change | Fast compilation check |
-| `cargo test --lib -p bubbaloop` | Before commits | 298+ Rust tests |
+| `cargo test --lib -p bubbaloop` | Before commits | 2400+ Rust tests |
 | `cargo test --features test-harness --test integration_mcp` | After MCP changes | 47 integration tests |
 | `pixi run clippy` | Before PRs | Zero warnings (`-D warnings`) |
 
@@ -107,7 +107,7 @@ See `CLAUDE.md` for full conventions. Critical rules:
 ## Pull Request Checklist
 
 - [ ] `pixi run check` passes
-- [ ] `cargo test --lib -p bubbaloop` (298+ tests)
+- [ ] `cargo test --lib -p bubbaloop` (2400+ tests)
 - [ ] `cargo test --features test-harness --test integration_mcp` (47 tests)
 - [ ] `pixi run clippy` zero warnings
 - [ ] `CLAUDE.md` updated if conventions changed

README.md

@@ -16,7 +16,7 @@ AI agents revolutionized software engineering. **Bubbaloop brings that same powe
 | **Data plane** | None | **Zenoh (zero-copy pub/sub)** |
 | **Hardware** | None | **Self-describing sensor nodes** |
 | **Runs on** | Desktop / cloud | **Jetson, RPi, any Linux ARM64/x86** |
-| **MCP role** | Client (consumes tools) | **Server (42+ tools, 3-tier RBAC)** |
+| **MCP role** | Client (consumes tools) | **Server (47 tools, 3-tier RBAC)** |
 | **Scheduling** | Always-on LLM (~$5-10/day) | **Offline Tier 1 + LLM Tier 2 (~$0.05/day)** |
 
 ## Quick Install
@@ -43,20 +43,29 @@ All run as systemd user services with autostart enabled.
 
 ## Login & Authentication
 
+Bubbaloop supports three providers — **Gemini** (recommended for new users), **Claude**, and **Ollama** (local).
+
 ```bash
-# Option 1: API Key (pay-as-you-go)
+# Option 1 (Recommended): Gemini — free tier, strong vision support
+export GEMINI_API_KEY=...           # from https://aistudio.google.com/apikey
+# or write the key to ~/.bubbaloop/gemini-key
+
+# Option 2: Claude API Key (pay-as-you-go)
 bubbaloop login
 # → Choose [1], paste your key from console.anthropic.com
 
-# Option 2: Claude Subscription (Pro/Max/Team)
+# Option 3: Claude Subscription (Pro/Max/Team)
 claude setup-token    # Run in Claude Code CLI first
 bubbaloop login
 # → Choose [2], paste the sk-ant-oat01-* token
 
-# Check auth status
+# Option 4: Ollama — fully local, no cloud bill
+ollama serve          # in a separate terminal
+
+# Check auth status (shows all configured providers)
 bubbaloop login --status
 
-# Remove credentials
+# Remove Claude credentials (Gemini/Ollama have no stored secret)
 bubbaloop logout
 ```
 
@@ -69,12 +78,15 @@ bubbaloop status
 # Start daemon (runs agent runtime + MCP server + node manager)
 bubbaloop up
 
-# Talk to your hardware via Claude AI (agents run daemon-side)
+# Talk to your hardware (Gemini by default; agents run daemon-side)
 bubbaloop agent chat "What sensors do I have?"
-bubbaloop agent chat                   # Interactive REPL
-bubbaloop agent chat -a camera-expert "describe the video feed"
+bubbaloop agent chat                   # Interactive REPL (Ctrl-C cancels the in-flight turn)
+bubbaloop agent chat -a camera-expert "describe what the terrace camera sees"
 bubbaloop agent list                   # Show running agents + models
 
+# Or open the dashboard's Chat tab in a browser — same agent runtime,
+# rich tool-call traces, and image-in-chat when the agent calls grab_frame.
+
 # System diagnostics with auto-fix
 bubbaloop doctor --fix
 
@@ -161,13 +173,13 @@ export BUBBALOOP_MCP_TOKEN=$(bubbaloop mcp --token)
 
 47 MCP tools + agent-internal tools. Configure Claude Code via `.mcp.json` (already in project root).
 
-**Agent-internal tools** (daemon-side only, not exposed via MCP): `memory_search`, `memory_forget`, `schedule_task`, `create_proposal`, `read_file`, `write_file`, `run_command`.
+**Agent-internal tools** (daemon-side only, not exposed via MCP): `grab_frame` (Zenoh-query a camera node for one JPEG → fed to a vision-capable provider), `memory_search`, `memory_forget`, `schedule_task`, `create_proposal`, `read_file`, `write_file`, `run_command`, `publish_to_topic`.
 
 ## Architecture
 
 ```
                     ┌──────────────────────────────────┐
-                    │   AI Agent (Claude via MCP)       │
+                    │   AI Agent (Gemini/Claude via MCP)│
                     │   http://127.0.0.1:8088/mcp      │
                     └──────────────┬───────────────────┘
                                    │

ROADMAP.md

@@ -63,7 +63,7 @@ bubbaloop agent chat   # interactive REPL
 │  │  World State | Short-term | Episodic | Semantic     │  │
 │  └──────────────────────┬─────────────────────────────┘  │
 │  ┌──────────────────────┴─────────────────────────────┐  │
-│  │  MCP Server (42 tools) — sole control interface     │  │
+│  │  MCP Server (47 tools) — sole control interface     │  │
 │  └──────────────────────┬─────────────────────────────┘  │
 │  ┌──────────────────────┴─────────────────────────────┐  │
 │  │  Daemon (skill runtime + agent host)                │  │
@@ -91,7 +91,7 @@ bubbaloop agent chat   # interactive REPL
 ### v0.0.1–v0.0.6: MCP-Native Sensor Runtime
 
 - [x] Single binary: CLI + daemon + MCP server
-- [x] 42 MCP tools (discovery, lifecycle, data, config, system, memory, telemetry, beliefs, missions, constraints, alerts)
+- [x] 47 MCP tools (discovery, lifecycle, data, config, system, memory, telemetry, beliefs, missions, constraints, alerts, toolsets)
 - [x] MCP is sole control interface — Zenoh for data only
 - [x] Marketplace with precompiled binaries (ARM64 + x86_64)
 - [x] Full node lifecycle via MCP: install, uninstall, start, stop, restart, autostart
@@ -163,7 +163,7 @@ config:
 **Deliverables:**
 - [x] `bubbaloop agent chat` CLI command (thin Zenoh client, LLM runs daemon-side)
 - [x] `bubbaloop agent list` — discover running agents via manifest queryables
-- [x] `bubbaloop agent setup` — interactive wizard: provider (Claude/Ollama), model, identity
+- [x] `bubbaloop agent setup` — interactive wizard: provider (Gemini/Claude/Ollama), model, identity
 - [x] Multi-agent runtime: agents run as tokio tasks inside daemon
 - [x] Agent gateway: Zenoh pub/sub wire format (inbox/outbox/manifest topics)
 - [x] Per-agent config via `~/.bubbaloop/agents.toml` (provider, model override, capabilities)
@@ -174,8 +174,11 @@ config:
 - [x] Internal MCP tool dispatch (call tools without HTTP round-trip)
 - [x] System prompt injection: sensor inventory, node status, active schedules
 - [x] First-run LLM onboarding: marker file triggers interview prompt; agent writes its own `identity.md`
+- [x] HTTP+SSE chat endpoint — dashboard Chat tab talks to daemon's agent gateway (PR #91)
+- [x] Gemini Vision provider — `ContentBlock::Image` + `grab_frame` agent tool (PR #96)
+- [x] Client-side turn cancellation via Zenoh `agent/{id}/cancel` topic (PR #94/95)
+- [x] Multi-provider login status (Gemini / Claude / Ollama) in `bubbaloop login --status` (PR #93)
 - [ ] Capability-based message routing (currently falls back to default agent)
-- [ ] HTTP chat endpoint for future dashboard integration
 
 **New deps:** None (`reqwest` already in dep tree). **New code:** ~500-800 lines.
 
@@ -373,7 +376,7 @@ Memory confirmed or contradicted by sensor readings — not just conversation hi
 | Runtime | Rust + Tokio | Memory safety, small binary, edge-ready |
 | Data plane | Zenoh | Zero-copy pub/sub, decentralized, Rust-native |
 | Schemas | Protobuf + prost | Self-describing, runtime introspection |
-| Control | MCP (rmcp) | Standard AI agent interface, 42 MCP tools + agent-internal tools |
+| Control | MCP (rmcp) | Standard AI agent interface, 47 MCP tools + ~43 agent-internal tools |
 | Memory | SQLite (rusqlite) | Embedded, +1-2 MB, battle-tested everywhere |
 | CLI | argh | Minimal, fast compile |
 | Logging | log + env_logger | Simple, stderr-only |