A fully local, governance-ready terminal — your AI, your agents, your rules, your machine.

Kairos is a fork of Warp (via OpenWarp) that strips all mandatory cloud dependencies, wires in specsmith as a local AI governance engine, and opens the AI provider layer to any OpenAI-compatible endpoint.

v0.1.0-alpha — Early access release. Not production-ready. Not affiliated with Warp, Inc.

---

What Kairos is

Upstream Warp requires a Warp account, Warp's cloud servers, and Warp's AI gateway. Kairos removes all of that and replaces it with a locally-governed stack:

	Upstream Warp	Kairos
Cloud dependency	Hard dependency on Warp backend	Zero — no account, no login, no cloud calls
AI governance	Warp cloud + opaque server rules	specsmith local governance engine
AI provider	Warp gateway only	Any OpenAI-compatible endpoint, BYOE
Default BYOE endpoint	warp.dev servers	http://127.0.0.1:7700 (local specsmith)
Credentials	Cloud account	Local config, never leaves the device
Bug reporting	Warp feedback form	GitHub Issues: kairos or specsmith repo
Blocks / Workflows / Keymaps	Kept	Fully preserved
Terminal core	Full Warp UX	Full Warp UX
License	AGPL-3.0 / MIT dual	Same (see LICENSE)

How it works

01 · specsmith governance-serve starts locally At launch, Kairos spawns specsmith governance-serve as a managed child process on port 7700. This is the local AI governance backend — preflight checks, verification, confidence scoring, and audit all run on your machine.

02 · BYOE wired to localhost by default The AI provider endpoint defaults to http://127.0.0.1:7700/v1/. Point it at any OpenAI-compatible endpoint in Settings → AI if you want a different model or provider. Credentials are stored locally only.

03 · Full Warp terminal experience, zero cloud Blocks, Workflows, AI agent sessions, themes, keymaps, SSH manager — all the Warp UX you know, running entirely offline without any server dependency.

Core features

• Local AI governance — specsmith runs preflight and verification checks locally before and after every governed action
• BYOE — any OpenAI Chat Completions-compatible endpoint works out of the box; 6 native protocols via genai
• Zero forced login — skip_login is always active; no Warp account required
• Zero telemetry — all analytics, crash uploads, and experiment flags disabled
• EU AI Act / NIST AI RMF compliance — cryptographic audit log, AI disclosure, human escalation, kill-switch, least-privilege permissions, and compliance export (see specsmith compliance docs)
• Governance Tools Panel — Settings → Governance surfaces live compliance controls, context window settings, permission profiles, and the kill-switch per session/project
• Per-project shell memory — shell preference is saved to .kairos/shell-pref.json so every new tab in a project opens the same shell you last chose
• Context window fill indicator — live progress bar in the agent footer shows current context fill percentage; auto-compression fires at 80%; hard 15% ceiling enforced
• Bug reporting via GitHub Issues — bug links in the Governance page open directly in the browser: terminal bugs → kairos, AI/governance bugs → specsmith
• Gruvbox Dark default theme — new users start with Gruvbox Dark instead of the Kairos Amber theme (still available in Settings → Themes)
• SSH Integration — block-based input and shell integration for SSH sessions (previously called "Warpify"; all user-visible strings updated)
• Full Warp UX preserved — Blocks, Workflows, AI commands, Keymaps, SSH manager, themes, split panes, MCP client — all kept and working

---

AI Compliance & Governance

Kairos is the UI surface for specsmith's full compliance stack. All compliance enforcement happens inside specsmith; Kairos surfaces the controls and status.

Governance Tools Panel

Settings → Governance is the central compliance dashboard for the active session:

• specsmith status — live version, health, and update availability (checks pipx → pip → pip3, with clickable update buttons)
• Permission profile — view and change the active least-privilege preset (read_only / standard / extended / admin) or set custom allow/deny lists
• Escalation threshold — the confidence level below which Kairos asks for human confirmation before executing a governed action
• Kill-switch — immediately terminates all active agent sessions and writes a kill event to the project LEDGER.md; satisfies EU AI Act Art. 14 §4
• Audit log viewer — live tail of .specsmith/trace.jsonl with chain integrity status (green = intact, red = tampered)
• Context window — Ollama num_ctx recommendation based on detected GPU VRAM, compression threshold, and auto-compress toggle
• Bug report links — clickable links that open the correct GitHub Issues repo in the system browser (terminal bugs vs. AI/governance bugs routed separately)

Compliance settings cascade: global defaults in ~/.specsmith/config.yml, then per-project .specsmith/config.yml, then per-session overrides in the panel. Changes in the panel can be written back to the per-project config.

Compliance Standards Met

Kairos + specsmith together implement:

EU AI Act (Regulation 2024/1689)

• Art. 9 — Risk Management: AEE verification loop with confidence scoring
• Art. 12 — Logging: SHA-256 chained TraceVault (tamper-evident, append-only)
• Art. 13 — Transparency: ai_disclosure in every preflight response
• Art. 14 — Human Oversight: escalation threshold + kill-switch
• Art. 15 — Robustness: bounded retry, confidence gates, hard context ceiling
• Art. 53 — GPAI Transparency: provider + model name in every AI response

NIST AI RMF 1.0

• GOVERN: H1–H13 governance rules, permission profiles, per-project policy
• MAP: AEE stress-test, belief graph, contradiction and uncertainty metrics
• MEASURE: confidence scoring, epistemic equilibrium, specsmith epistemic-audit
• MANAGE: kill-switch, escalation, bounded retry, safe-write backup, deny-list guardrails

See specsmith's compliance documentation for the full breakdown of each mechanism.

---

Per-Project Shell Memory

Kairos remembers the shell you use in each project. When you open a new tab with an explicit shell (right-click → New Tab With Shell), that choice is saved to .kairos/shell-pref.json at the project root:

{ "shell": { "WSL": "Ubuntu-24.04" } }

Subsequent new tabs in the same project automatically open that shell — no global setting is changed. The project root is detected by walking up from the current directory until .git, .kairos, or scaffold.yml is found.

Supported shell variants: system default, executable path, WSL distro, MSYS2, custom.

---

Context Window Management

Kairos shows real-time context usage and handles overflow automatically via specsmith.

Fill indicator — a compact progress bar in the agent footer shows current fill as a percentage. Color transitions: green → yellow at 80%, red at the hard ceiling.

Auto-compression — when fill reaches the compression threshold (default 80%), Kairos fires SummarizeAIConversation before the next agent turn. The conversation is condensed to a summary that preserves key decisions and context.

Hard ceiling — a 15% reservation (minimum 2,048 tokens) is always held back. Fill cannot reach 100% — if it would, ContextFullError is raised and emergency compression runs first. This is a safety invariant, not a setting.

GPU-aware sizing — the Governance panel shows the recommended num_ctx for your Ollama setup based on detected NVIDIA or AMD GPU VRAM:

VRAM	Recommended Context
< 6 GB	4,096 tokens
6–11 GB	8,192 tokens
12–19 GB	16,384 tokens
20 GB+	32,768 tokens

Verified AI providers

Provider	Base URL	Notes
specsmith (local)	http://127.0.0.1:7700/v1	Default — governance-aware local endpoint
OpenAI	https://api.openai.com/v1	Direct
Anthropic	via genai native	Claude family
DeepSeek	https://api.deepseek.com/v1	thinking + tool calling
Gemini	via genai native	Google AI Studio
Ollama	http://localhost:11434/v1	Local inference, no key
OpenRouter	https://openrouter.ai/api/v1	Aggregator gateway

Build from source

Requires: Rust 1.92+ (via rustup), protoc (for proto API crates).

git clone https://github.com/BitConcepts/kairos
cd kairos
./script/bootstrap   # platform-specific deps (macOS/Linux)
./script/run         # build & run
./script/presubmit   # fmt / clippy / tests

Windows: install deps via winget first:

winget install Rustlang.Rustup
winget install Google.Protobuf

Always target the kairos binary explicitly:

cargo build --release --bin kairos
cargo run   --release --bin kairos

Windows shortcut — a convenience script is included at the repo root:

.\Open-Kairos.ps1          # debug build + launch
.\Open-Kairos.ps1 -Release # release build + launch
.\Open-Kairos.ps1 -NoBuild # skip build, run last compiled binary

Do not run cargo build --release --bin {warp,stable,dev,preview} — those entry points require Warp's private warp-channel-config binary and will panic at startup. Use kairos only.

See DEVELOPMENT.md for the full engineering guide.

License

Kairos inherits the dual-license structure from Warp. See LICENSE for the full breakdown:

• crates/warpui_core / crates/warpui — MIT (Copyright 2020-2026 Denver Technologies, Inc.)
• All other upstream Warp code — AGPL-3.0 (Copyright 2020-2026 Denver Technologies, Inc.)
• Kairos-specific additions (crates/kairos-governance, specs/, .github/, themes/kairos_amber.yaml, and files with a BitConcepts copyright header) — MIT (Copyright 2026 BitConcepts)

Contributing

Community contributions welcome. See CONTRIBUTING.md for the full flow.

Before filing, please search existing issues. Security vulnerabilities should be reported privately per CONTRIBUTING.md#reporting-security-issues.

Acknowledgements

Kairos stands on the shoulders of many teams and open-source projects:

Core foundation Warp — the terminal this is forked from, built by Warp, Inc. OpenWarp — the community fork that first removed cloud dependencies, laying the groundwork for Kairos.

Kairos governance specsmith — the local AI governance engine powering Kairos.

Key dependencies genai · Tokio · minijinja · cosmic-text · Alacritty VTE · Hyper · reqwest · wgpu