# Bote Roadmap
> **Principle**: One protocol implementation for the entire ecosystem. Tools are registered, not reimplemented.
Completed items are in [CHANGELOG.md](../../CHANGELOG.md).
---
## v0.23.3 — TypeScript Bridge (done)
### TypeScript bridge
- [x] HTTP bridge server — Rust bote core exposes tools via HTTP, SY's TypeScript calls it
- [x] Tool result serialization compatible with SY's MCP format
- [x] Health endpoint for bridge liveness
### Cross-node discovery
- [x] Cross-node tool discovery via majra
---
## v0.24.3 — Advanced Registry (done)
### Schema validation
- [x] Full JSON Schema validation (not just required fields)
- [x] Type checking (string, number, boolean, array, object)
- [x] Enum constraints
- [x] Default values
### Tool versioning
- [x] Version field in ToolDef
- [x] Capability negotiation — client requests version, server matches
- [x] Deprecation warnings for old tool versions
### Dynamic registration
- [x] Runtime tool registration/deregistration
- [x] Hot-reload tool handlers without restart
- [x] Tool namespacing (project_tool format enforcement)
---
## v0.25.3 — Consumer Integration
### Adoption
- [x] daimon uses bote host types + real MCP dispatch (external tool forwarding via reqwest, built-in handler table)
- [x] 19 consumer apps using bote (daimon, agnoshi, agnosai, t-ron, jalwa, nein, stiva, itihas, varna, selah, hoosh, vidya, rasayan, szal, tarang, vidhana, nazar, mneme, tazama)
- [x] SY integrates bote via NAPI bindings (bridge deprecated — SY migrating to full Rust)
- [x] agnoshi uses bote JSON-RPC types for native MCP dispatch (replaced curl subprocess with reqwest + bote protocol)
### Validation
- [x] Cross-crate integration tests (12 tests covering daimon, agnoshi, t-ron, consumer patterns)
- [x] Protocol conformance test suite
- [x] Performance: dispatch overhead < 1µs per tool call (measured ~195ns)
---
## v1.0.0 Criteria
- [x] JSON-RPC 2.0 fully compliant (batch, notifications, streaming)
- [x] All 3 transports stable (stdio, HTTP, WebSocket)
- [x] libro audit integration for every tool call
- [x] 3+ downstream consumers in production (19 repos integrated)
- [x] 90%+ test coverage (90.65% line coverage)
- [x] docs.rs complete
- [x] Protocol conformance test suite passing
- [x] Dispatch benchmark: < 500ns per tool call
---
## Post-v1
### Advanced
- [ ] gRPC transport (proto definitions for MCP)
- [x] Tool sandboxing via kavach (execute untrusted tool handlers in isolation)
- [ ] Tool marketplace — discover and install tools from mela
- [ ] AI-powered tool selection — hoosh suggests which tool to call based on intent
- [ ] Tool composition — chain multiple tools into a workflow via szál
### Ecosystem
- [ ] jnana knowledge tools — `jnana_search`, `jnana_lookup`, `jnana_budget` registered as bote tools for MCP-accessible knowledge queries
### Platform
- [ ] WASM tool handlers (run tool logic in wasmtime)
- [ ] Python tool handlers (PyO3 bridge)
- [ ] C FFI for external tool integration
---
## Non-goals
- **Tool implementation** — bote dispatches to handlers, doesn't implement business logic
- **LLM integration** — that's hoosh. Bote doesn't decide which tool to call
- **Workflow orchestration** — that's szál. Bote calls one tool at a time
- **Agent lifecycle** — that's daimon/AgnosAI. Bote doesn't manage agents