Tau is a Rust-first agent runtime and operator control plane with a connected core path: CLI runtime -> sessions/tools/safety -> gateway APIs -> transport and operator workflows.
Short answer to the integration question: the core runtime path is integrated and runnable today, while some subsystems are still staged or partial.
Tau combines:
- a primary CLI runtime (
tau-coding-agent) for interactive and one-shot execution, - persistent session and tool-policy/safety controls,
- gateway and operator surfaces for API and operations workflows,
- deterministic demos and validation scripts for local and CI loops.
The workspace is intentionally multi-crate and contract-driven. Full crate membership is in Cargo.toml.
- Operators who need repeatable runtime controls, diagnostics, and rollback-friendly workflows.
- Integrators who need OpenAI-compatible gateway routes and transport bridges.
- Contributors working in a spec-driven, TDD-oriented Rust workspace.
These are the paths that operate as connected flows today.
| Path | Start Point | Integrated Components | Primary Evidence |
|---|---|---|---|
| Local operator loop | cargo run -p tau-coding-agent -- --onboard --onboard-non-interactive then prompt mode |
CLI runtime, agent core loop, sessions, tools, safety policies | docs/guides/quickstart.md, docs/guides/operator-control-summary.md |
| Gateway auth/session loop | ./scripts/demo/gateway-auth-session.sh |
Gateway auth/session handling, API route contracts, runtime policies | docs/guides/gateway-auth-session-smoke.md, docs/guides/gateway-api-reference.md |
| Unified runtime lifecycle loop | ./scripts/run/tau-unified.sh up --auth-mode localhost-dev |
One-command runtime bring-up (up/status/down) for gateway/dashboard + interactive TUI agent (tui) with explicit live-shell fallback |
scripts/run/tau-unified.sh, docs/guides/operator-deployment-guide.md |
| Multi-channel ingress loop | ./scripts/demo/multi-channel.sh |
Multi-channel runtime, transport normalization, routing pipeline | docs/guides/multi-channel-event-pipeline.md, docs/guides/transports.md |
| Prompt optimization loop | docs/guides/training-ops.md runbook flow |
Training runner/store/tracer/proxy + rollout controls | docs/guides/training-ops.md, docs/guides/training-proxy-ops.md |
| Connected operator GA loop | ./scripts/verify/m296-ga-readiness-gate.sh |
RL maturity wave + auth/readiness checks + rollback trigger validation + closeout signoff criteria | docs/guides/m296-ga-readiness-gate.md, artifacts/operator-ga-readiness/verification-report.json |
- Run interactive and one-shot agent flows from
tau-coding-agent. - Use session persistence and lifecycle operations (branch, resume, export/import/repair).
- Route model calls across multiple provider/auth modes.
- Run gateway API surfaces and operator routes.
- Use built-in tools with policy controls (filesystem/shell/http/path/rate/sandbox).
- Run channel and bridge runtimes (GitHub Issues, Slack, Discord, Telegram/WhatsApp paths).
- Operate prompt-optimization workflows with SQLite-backed rollout state and optional proxy attribution.
- Execute deterministic demo suites and validation scripts in local/CI loops.
Some surfaces are intentionally diagnostics-first or staged:
- True RL:
- deterministic end-to-end harness is available (
tau-trainer), - policy-operations depth (promotion/rollback/significance/runtime-audit) is aggregated in
scripts/verify/m310-rl-policy-ops-depth.sh, - operations drill depth (operational safety, resume-after-crash, benchmark/safety, rollback checklist) is aggregated in
scripts/verify/m316-rl-operations-drill-depth.sh, - broader production policy-optimization operating loops remain an expansion track.
- deterministic end-to-end harness is available (
- Dashboard:
- route and diagnostics surfaces exist,
- operator workflow depth across ops chat/session/lineage/memory-graph/tools routes is aggregated in
scripts/verify/m314-dashboard-operator-workflow-depth.sh, - command-center depth (timeline/alert/control markers + control-action fail paths + live stream matrix) is aggregated in
scripts/verify/m318-dashboard-command-center-depth.sh, - not all desired product UX workflows are fully integrated live-mutation paths.
- Auth verification:
- provider auth-mode matrix and gateway auth/session lifecycle conformance are covered in deterministic suites (
scripts/verify/m295-operator-maturity-wave.sh), - auth-depth lifecycle and edge-path coverage is aggregated in
scripts/verify/m303-auth-workflow-depth.sh, - credential lifecycle depth (integration-auth set/status/rotate/revoke + resolve-secret fail-closed) is aggregated in
scripts/verify/m309-auth-credential-lifecycle-depth.sh, - live-env validation depth (skip/enable/key contracts + aggregated auth-depth gates) is aggregated in
scripts/verify/m312-auth-live-env-depth.sh, - live third-party credential/network validation remains environment-specific.
- provider auth-mode matrix and gateway auth/session lifecycle conformance are covered in deterministic suites (
- Multi-channel orchestration:
- deterministic C5 scenario-depth coverage is aggregated in
scripts/verify/m307-multi-channel-orchestration-depth.sh, - live connector/provider uptime and credential validation remain environment-specific.
- deterministic C5 scenario-depth coverage is aggregated in
- E2E core verification:
- deterministic integration+gateway core scenario depth is aggregated in
scripts/verify/m313-e2e-core-scenario-depth.sh, - deterministic operator-route scenario depth is aggregated in
scripts/verify/m315-e2e-operator-route-depth.sh, - full PRD-wide scenario-group completion continues as an expansion track.
- deterministic integration+gateway core scenario depth is aggregated in
- TUI:
- includes operator-shell, interactive
agent, and state-backedshell-livemodes, shell-livenow reports deterministic malformed/missing artifact diagnostics for operator triage,- operator workflow depth (shell, shell-live watch, and artifact diagnostics) is aggregated in
scripts/verify/m311-tui-operator-workflow-depth.sh, - scenario-expansion depth (demo mode behavior + parser/shell-live edge paths + workflow-depth chaining) is aggregated in
scripts/verify/m317-tui-scenario-expansion-depth.sh, - remains complementary to (not a replacement for) web dashboard workflows.
- includes operator-shell, interactive
| Capability Area | Status | Meaning | Primary Reference |
|---|---|---|---|
| Core CLI runtime + sessions + tools | Integrated | Production-like operating loop available | docs/guides/quickstart.md |
| Gateway auth/session APIs | Integrated | Deterministic auth/session smoke flows and documented API contracts | docs/guides/gateway-auth-session-smoke.md |
| Auth workflow conformance | Integrated | Provider matrix + gateway session lifecycle validated by dedicated suites | crates/tau-provider/tests/auth_workflow_conformance.rs, scripts/verify/m295-operator-maturity-wave.sh |
| Multi-channel and bridge transports | Operational | Runnable with connector-specific maturity differences | docs/guides/transports.md |
| Dashboard operator UX | Partial | Ops routes and diagnostics available; broader UX still expanding | docs/guides/dashboard-ops.md |
| Prompt optimization training | Integrated | Canonical training path today | docs/guides/training-ops.md |
| True RL | Integrated | Deterministic end-to-end harness emits rollout + GAE/PPO artifact evidence | crates/tau-trainer/src/rl_e2e.rs, crates/tau-trainer/src/bin/rl_e2e_harness.rs |
| TUI | Integrated | Operator-shell + interactive agent mode + state-backed shell-live diagnostics |
crates/tau-tui/src/main.rs, crates/tau-tui/src/lib.rs, scripts/verify/m295-operator-maturity-wave.sh |
| Gap | Current State | Execution Plan Links |
|---|---|---|
| True RL productionization depth | deterministic end-to-end harness, promotion/rollback gate verification, GA readiness gate, policy-operations depth verification, and operations drill-depth verification are delivered; larger-scale policy operations still expanding | docs/planning/integration-gap-closure-plan.md, docs/planning/true-rl-roadmap-skeleton.md, docs/guides/training-ops.md, scripts/verify/m301-rl-promotion-rollback-gate.sh, scripts/verify/m310-rl-policy-ops-depth.sh, scripts/verify/m316-rl-operations-drill-depth.sh, scripts/verify/m296-ga-readiness-gate.sh |
| Dashboard maturity expansion | deterministic live mutation depth, operator workflow-depth verification, and command-center depth verification now aggregate status/action/stream plus ops chat/session/lineage/memory-graph/tools/timeline/alert/control contracts; richer workflow UX still expanding | docs/planning/integration-gap-closure-plan.md, docs/guides/dashboard-ops.md, docs/guides/operator-deployment-guide.md, scripts/verify/m308-dashboard-live-mutation-depth.sh, scripts/verify/m314-dashboard-operator-workflow-depth.sh, scripts/verify/m318-dashboard-command-center-depth.sh |
| Multi-channel orchestration depth | deterministic C5 Telegram/Discord routing, WhatsApp webhook verification, lifecycle, and media-handling coverage is aggregated in a dedicated gate; live provider-specific behavior still environment-bound | docs/planning/integration-gap-closure-plan.md, docs/guides/transports.md, scripts/verify/m307-multi-channel-orchestration-depth.sh |
| Extended auth live-env verification | deterministic matrix/lifecycle coverage, auth-depth edge-path gating, credential lifecycle depth verification, and live-env validation depth gating are delivered; external credential/live-env permutations remain environment-specific | docs/planning/integration-gap-closure-plan.md, docs/provider-auth/provider-auth-capability-matrix.md, docs/guides/gateway-auth-session-smoke.md, scripts/verify/m303-auth-workflow-depth.sh, scripts/verify/m309-auth-credential-lifecycle-depth.sh, scripts/verify/m312-auth-live-env-depth.sh, scripts/verify/m296-live-auth-validation.sh |
| E2E scenario-group expansion | deterministic core scenario-depth and operator-route scenario-depth verification across integration memory/tool paths and gateway lifecycle/session/operator-route contracts are delivered; full scenario-group completion remains in progress | docs/planning/integration-gap-closure-plan.md, specs/milestones/m298/index.md, scripts/verify/m313-e2e-core-scenario-depth.sh, scripts/verify/m315-e2e-operator-route-depth.sh |
| TUI interaction depth | operator shell, interactive agent handoff, resilient shell-live diagnostics, deterministic workflow-depth verification, and deterministic scenario-expansion verification are delivered; richer UX flows continue to evolve |
docs/planning/integration-gap-closure-plan.md, crates/tau-tui, docs/guides/demo-index.md, scripts/verify/m311-tui-operator-workflow-depth.sh, scripts/verify/m317-tui-scenario-expansion-depth.sh |
Run commands from repository root.
- Prerequisite
rustup default stable- Fast validation loop
./scripts/dev/fast-validate.sh- Initialize local Tau state
cargo run -p tau-coding-agent -- --onboard --onboard-non-interactive- Run your first prompt
cargo run -p tau-coding-agent -- --prompt "Summarize src/lib.rs"- Optional TUI operator shell
cargo run -p tau-tui -- shell --width 88 --profile local-dev --no-color- Optional interactive TUI agent mode from runtime artifacts
cargo run -p tau-tui -- agent \
--dashboard-state-dir .tau/dashboard \
--gateway-state-dir .tau/gateway \
--request-timeout-ms 45000 \
--agent-request-max-retries 0 \
--width 88 \
--profile local-dev \
--no-color- Optional live TUI watch mode (read-only, multi-cycle refresh)
cargo run -p tau-tui -- shell-live --state-dir .tau/dashboard --width 88 --profile local-dev --watch --iterations 3 --interval-ms 1000 --no-colorFor a deeper walkthrough, use docs/guides/quickstart.md.
Fast local compile-focused loop:
./scripts/dev/fast-validate.sh --check-only --direct-packages-only --skip-fmtUnified one-command runtime entrypoint:
./scripts/run/tau-unified.sh up --auth-mode localhost-dev
./scripts/run/tau-unified.sh status
./scripts/run/tau-unified.sh tui --no-color
./scripts/run/tau-unified.sh tui --request-timeout-ms 90000 --agent-request-max-retries 1 --no-color
./scripts/run/tau-unified.sh tui --live-shell --iterations 3 --interval-ms 1000 --no-color
./scripts/run/tau-unified.sh downtau-unified.sh tui defaults to fast-fail interactive policy:
--request-timeout-ms 45000--agent-request-max-retries 0
Override defaults with flags above or env vars:
TAU_UNIFIED_TUI_REQUEST_TIMEOUT_MSTAU_UNIFIED_TUI_AGENT_REQUEST_MAX_RETRIES
Interactive TTY turns now emit progress markers to stderr while requests are
in-flight:
interactive.turn=start turn_timeout_ms=... request_timeout_ms=...interactive.turn=running elapsed_ms=...interactive.turn=end status=... elapsed_ms=...
Full pre-merge gate:
./scripts/dev/fast-validate.sh --fullInteractive runtime mode:
cargo run -p tau-coding-agent -- --model openai/gpt-5.2Gateway auth/session smoke:
./scripts/demo/gateway-auth-session.shDashboard demo path:
./scripts/demo/dashboard.shDemo index and selective runs:
./scripts/demo/index.sh --list
./scripts/demo/index.sh --only onboarding,gateway-auth,gateway-remote-access --fail-fast./scripts/demo/all.sh --list
./scripts/demo/all.sh --only local,rpc,events --fail-fastRL end-to-end deterministic harness:
cargo run -p tau-trainer --bin rl_e2e_harness -- --run-id local --output-dir artifacts/rl-e2e --print-jsonOperator maturity wave verification (TUI + RL + auth):
./scripts/verify/m295-operator-maturity-wave.shTUI interactive agent loop from runtime artifacts:
cargo run -p tau-tui -- agent \
--dashboard-state-dir .tau/dashboard \
--gateway-state-dir .tau/gateway \
--request-timeout-ms 45000 \
--agent-request-max-retries 0 \
--profile local-dev \
--no-colorTUI live watch loop from dashboard artifacts:
cargo run -p tau-tui -- shell-live --state-dir .tau/dashboard --profile local-dev --watch --iterations 3 --interval-ms 1000 --no-colorM296 GA readiness gate (Connected operator GA loop):
./scripts/verify/m296-ga-readiness-gate.shClean generated local artifacts:
./scripts/dev/clean-local-artifacts.shChecked-in example assets and starter package references:
./examples/starter/package.json./examples/extensions./examples/extensions/issue-assistant/extension.json./examples/extensions/issue-assistant/payload.json./examples/events./examples/events-state.json
See ./examples/README.md for package and asset walkthrough details.
Primary docs index: docs/README.md
Operator deployment guide: docs/guides/operator-deployment-guide.md
Gateway API reference (70+ routes): docs/guides/gateway-api-reference.md
Contributor guide: CONTRIBUTING.md
Security policy: SECURITY.md
Operator runbooks:
docs/guides/operator-deployment-guide.mddocs/guides/operator-control-summary.mddocs/guides/dashboard-ops.mddocs/guides/gateway-ops.mddocs/guides/memory-ops.mddocs/guides/m296-ga-readiness-gate.md
Integrator/API references:
docs/guides/gateway-api-reference.mddocs/guides/transports.mddocs/provider-auth/provider-auth-capability-matrix.md
Contributor references:
CONTRIBUTING.mdAGENTS.mddocs/tau-coding-agent/code-map.mddocs/architecture/crate-dependency-diagram.mddocs/guides/startup-di-pipeline.mddocs/guides/contract-pattern-lifecycle.mddocs/guides/multi-channel-event-pipeline.mddocs/guides/doc-density-scorecard.md
Planning and gap closure:
docs/planning/true-rl-roadmap-skeleton.mddocs/guides/roadmap-execution-index.mddocs/planning/integration-gap-closure-plan.md
Core runtime:
crates/tau-coding-agentcrates/tau-agent-corecrates/tau-runtimecrates/tau-orchestrator
Gateway and ops:
crates/tau-gatewaycrates/tau-dashboardcrates/tau-dashboard-uicrates/tau-ops
Model and policy:
crates/tau-aicrates/tau-providercrates/tau-toolscrates/tau-safety
State and extension surfaces:
crates/tau-sessioncrates/tau-memorycrates/tau-extensionscrates/tau-skills
Transport/bridge runtimes:
crates/tau-github-issues-runtimecrates/tau-slack-runtimecrates/tau-discord-runtimecrates/tau-multi-channel
Training and algorithms:
crates/tau-training-typescrates/tau-training-storecrates/tau-training-tracercrates/tau-training-runnercrates/tau-training-proxycrates/tau-trainercrates/tau-algorithm
Local Docker smoke build:
./scripts/dev/docker-image-smoke.sh --tag tau-coding-agent:local-smokeRelease workflow and artifacts:
.github/workflows/release.ymldocs/guides/release-automation-ops.md- GitHub Releases: https://github.com/njfio/Tau/releases
- Security reporting policy:
SECURITY.md - Contribution guide:
CONTRIBUTING.md - Issue/spec workflow contract:
AGENTS.md
This repository expects issue-first, spec-driven, test-driven changes with explicit validation evidence.