MIRA v0.318.7 — public beta

MIRA v0.318.7 — public beta

MIRA is a self-hostable AI assistant that runs as a single binary. Published by Vexillon under AGPL-3.0-or-later.

Public beta. Supersedes v0.318.0. A stabilization + context-efficiency series (0.318.1–0.318.7) focused on running reliably against local models. No new configuration and no breaking changes — safe to upgrade in place.

Leaner prompts on local models

• Up to ~83% fewer prompt tokens per turn. With prompt caching and adaptive tool selection on, MIRA now ships a small, stable core toolset plus on-demand tool search instead of every tool's schema on every turn. In a measured chat turn this dropped the prompt from ~24K to ~4K tokens — cheaper, faster, and far easier for a small local model to handle.
• Context budgeted to what the model can actually use. MIRA sizes the working window to a model's reliable-recall range rather than an inflated advertised maximum, keeps injected memory/wiki context to a modest share of that window (prioritizing curated wiki), and now includes the tool-schema block in its per-turn token estimate (previously undercounted by ~10x).
• Startup context sanity check. On start, MIRA best-effort checks your configured context length against the model's actually-loaded window — for both LM Studio and Ollama — and warns when they don't match (a common cause of silent truncation with Ollama's small default context).

Reliability & correctness

• Degeneracy guard now also watches the reasoning stream. A reasoning model that gets stuck repeating itself while "thinking" is now caught and the turn is failed early, instead of running to the token cap.
• No more HTTP 400 from stacked system messages. MIRA coalesces system content into a single leading block before sending, so local chat templates that reject multiple or non-leading system messages work correctly.
• Config saves are atomic. Writes go through a temp file with fsync + rename and a backup, so an interrupted or racing save can no longer truncate your live config.
• Large tool results can't overflow a small model — they're clamped to the effective window mid-conversation.
• Per-turn model overrides are covered by the degeneracy guard (previously they bypassed it).
• UI-set provider keys win over a stale .env. A key set in the app is no longer silently overridden by an environment value; .env now only fills an empty setting, with a loud warning when it would shadow a configured key.

Install

# Linux / macOS
curl -fsSL https://get.vexillon.ai/install.sh | sh

# Windows
irm https://get.vexillon.ai/install.ps1 | iex

Targets

• x86_64-unknown-linux-gnu
• aarch64-unknown-linux-gnu
• x86_64-apple-darwin
• aarch64-apple-darwin
• x86_64-pc-windows-msvc

Verify

All artifacts are signed with minisign. Verify the checksum file, then the archive checksums:

minisign -Vm SHA256SUMS -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO
sha256sum -c SHA256SUMS

Trusted comment on a valid signature reads: signed by vexillon.ai — MIRA 0.318.7

MIRA v0.318.0 — public beta

MIRA v0.318.0 — public beta

MIRA is a self-hostable AI assistant that runs as a single binary. Published by Vexillon under AGPL-3.0-or-later.

Public beta. Supersedes v0.316.1. This release rolls up the changes from v0.317.0 as well (the in-app config panel for apps). No breaking changes — safe to upgrade in place.

Apps framework — sturdier HTTP tools

• Base URLs work with or without a trailing slash. ${config.base_url}/... template joins now collapse a redundant slash, so an app configured against, say, a Home Assistant instance no longer 404s on double-slash paths.
• Large responses can't blow the context. An app's HTTP-tool response is now clipped to a small model-facing cap (~24 KB) before it reaches the model.
• New optional response projection. An app can declaratively select a few dotted fields and a row limit to reduce a large JSON reply to a compact list — no code execution involved.
• Plus the in-app config panel for apps, folded in from v0.317.0.

Reliability fixes for local models

• Right-sized context window for local providers. New agent.local_effective_context_tokens (default 8192). When a local provider doesn't declare context_length_tokens, MIRA now budgets to this small, reliable window instead of trusting an advertised 128K that quantized local models don't actually attend to — which keeps prompts in the model's reliable zone and turns on compaction. An explicit context_length_tokens still takes precedence.
• Degenerate-output guard (on by default). New agent.degeneracy_guard.* settings. A guard wraps the provider chain and aborts pathologically repetitive output early, treats the turn as failed (not persisted, extractors skipped), and counts it as a provider error so fallback can kick in.
• Guardian triage model falls back sensibly. An empty guardian.triage_model now resolves to the provider's configured default model instead of sending an empty model name.
• Verified and closed the Just-in-Time Tools find_tools hint (shipped in v0.316.1).

Install

# Linux / macOS
curl -fsSL https://get.vexillon.ai/install.sh | sh

# Windows
irm https://get.vexillon.ai/install.ps1 | iex

Targets

• x86_64-unknown-linux-gnu
• aarch64-unknown-linux-gnu
• x86_64-apple-darwin
• aarch64-apple-darwin
• x86_64-pc-windows-msvc

Verify

All artifacts are signed with minisign. Verify the checksum file, then the archive checksums:

minisign -Vm SHA256SUMS -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO
sha256sum -c SHA256SUMS

Trusted comment on a valid signature reads: signed by vexillon.ai — MIRA 0.318.0

MIRA v0.316.1 — public beta

MIRA v0.316.1 — public beta

MIRA is a self-hostable AI assistant that runs as a single binary. Published by Vexillon under AGPL-3.0-or-later.

Public beta. This is a patch release and supersedes v0.316.0 — two bug fixes, no new features and no breaking changes. Safe to upgrade in place.

Fixes

• Guardian: no more bogus channel-restart proposals. The Signal "no inbound in 72h" health check no longer escalates to Red. Inbound silence tracks your activity, not daemon health, so a genuinely quiet channel was tripping the alarm and driving the Guardian to repeatedly propose pointless channel restarts. It's now at most an informational Yellow that points you at the real liveness signal. The Guardian also now sees the actual channel account IDs and validates a restart target against them before proposing — so it can no longer suggest restarting an account that doesn't exist.
• Just-in-Time Tools: MIRA stops claiming it lacks a tool it has. With adaptive tool selection enabled, MIRA now tells the model how many tools are currently loaded versus how many more it can pull in on demand, and to search for a tool before telling you a capability is unavailable. Previously that hint was missing, so a tool that wasn't loaded yet could be wrongly reported as unavailable.

Install

# Linux / macOS
curl -fsSL https://get.vexillon.ai/install.sh | sh

# Windows
irm https://get.vexillon.ai/install.ps1 | iex

Targets

• x86_64-unknown-linux-gnu
• aarch64-unknown-linux-gnu
• x86_64-apple-darwin
• aarch64-apple-darwin
• x86_64-pc-windows-msvc

Verify

All artifacts are signed with minisign. Verify the checksum file, then the archive checksums:

minisign -Vm SHA256SUMS -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO
sha256sum -c SHA256SUMS

Trusted comment on a valid signature reads: signed by vexillon.ai — MIRA 0.316.1

MIRA v0.316.0 — public beta

MIRA v0.316.0 — built web apps work behind a reverse proxy

A self-hostable, single-binary personal AI agent that runs on hardware you control. A project of Vexillon. Licensed under AGPL-3.0-or-later.

⚠️ Public beta (pre-release). Recommended build for new installs; supersedes v0.315.2.

Open your built web apps from anywhere — including behind a reverse proxy

When MIRA's coding agent builds a runnable web app or game, MIRA now serves it from its main URL at <your MIRA address>/a/<app-id>/. Because that's the same address you already expose, built apps now open correctly behind a reverse proxy (Traefik, Caddy, nginx) or tailscale serve — deployments where a separate app port isn't reachable — with no extra proxy configuration. They keep working on direct LAN and from the mobile app too.

Untrusted, model-generated app code stays isolated: the app runs inside a sandboxed frame with its own opaque origin, so it can't read your MIRA session or call the API even though it's served from the same host. The link MIRA hands you is now this reliably-reachable one, rather than an address that only worked on the same machine.

Under the hood

• CI now pins an exact Rust compiler version for reproducible builds across platforms.

Install

Linux / macOS: curl -fsSL https://get.vexillon.ai/install.sh | sh · Windows: irm https://get.vexillon.ai/install.ps1 | iex

Or download the signed build for your platform below, verify it against the vexillon.ai release key (below), extract the mira binary onto your PATH, and run mira setup. The web UI is bundled into the binary.

Targets

mira-0.316.0-x86_64-unknown-linux-gnu.tar.gz · -aarch64-unknown-linux-gnu.tar.gz · -x86_64-apple-darwin.tar.gz · -aarch64-apple-darwin.tar.gz · -x86_64-pc-windows-msvc.zip

Verify

Signed with minisign under vexillon.ai:

sha256sum -c SHA256SUMS
minisign -Vm mira-0.316.0-x86_64-unknown-linux-gnu.tar.gz \
  -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO

Trusted comment: signed by vexillon.ai — MIRA 0.316.0.

MIRA v0.315.2 — public beta

MIRA v0.315.2 — reliable restart after self-update

A self-hostable, single-binary personal AI agent that runs on hardware you control. A project of Vexillon. Licensed under AGPL-3.0-or-later.

⚠️ Public beta (pre-release). Recommended build for new installs; supersedes v0.315.1.

A focused reliability fix for a crash-loop some users could hit after an in-place self-update (most visible on Windows).

What was wrong

After updating in place, MIRA could fail to come back — the service sat stopped and manual starts looped, unable to bind its ports. The cause: a helper process a task had spawned held on to MIRA's network ports after MIRA exited, so the new instance couldn't start.

Fixed

• MIRA's listening ports are no longer inheritable by child processes, so a spawned helper can never keep a port bound after MIRA exits. This is the fix that makes self-update reliable.
• Background task helpers are fully cleaned up when their task ends — no lingering processes squatting ports.
• Clearer failures + deterministic restart: if a port really is in use, MIRA now exits with a plain message telling you which port and how to free it, instead of looping silently; and an update-triggered restart relaunches deterministically.

Note

The fix is active once you're running 0.315.2 — the update into this version still uses the previous restart path. If that one hiccups, free the port it names and start MIRA once; from here on it's fixed.

Install

Linux / macOS: curl -fsSL https://get.vexillon.ai/install.sh | sh · Windows: irm https://get.vexillon.ai/install.ps1 | iex

Or download the signed build for your platform below, verify it against the vexillon.ai release key (below), extract the mira binary onto your PATH, and run mira setup. The web UI is bundled into the binary.

Targets

mira-0.315.2-x86_64-unknown-linux-gnu.tar.gz · -aarch64-unknown-linux-gnu.tar.gz · -x86_64-apple-darwin.tar.gz · -aarch64-apple-darwin.tar.gz · -x86_64-pc-windows-msvc.zip

Verify

Signed with minisign under vexillon.ai:

sha256sum -c SHA256SUMS
minisign -Vm mira-0.315.2-x86_64-unknown-linux-gnu.tar.gz \
  -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO

Trusted comment: signed by vexillon.ai — MIRA 0.315.2.

MIRA v0.315.1 — public beta

MIRA v0.315.1 — web apps on mobile, real charts, friendlier links

A self-hostable, single-binary personal AI agent that runs on hardware you control. A project of Vexillon. Licensed under AGPL-3.0-or-later.

⚠️ Public beta (pre-release). Recommended build for new installs; supersedes v0.313.0.

Web apps you can actually open — including on mobile

• When MIRA builds a runnable web app or game, it now hands you a friendly, clickable link (e.g. "Snake", from the app's page title) instead of a raw URL.
• Built web apps are now reachable from the mobile app and across your LAN, not just a browser on the same machine — MIRA serves each app both at its per-app subdomain and at a network-reachable port path. (server.web_apps.mode now defaults to both; set it to subdomain if you only open apps on the MIRA box.)

Real, professional data charts

• Ask for a pie/bar/line chart or a BI graph and MIRA now produces an accurate matplotlib chart rendered inline (via its code sandbox), instead of ASCII art or a generative-image guess. If the charting backend isn't set up it tells you exactly what to enable rather than failing silently. (Requires the scientific sandbox backend, sandbox.pyodide.enabled.)

Reliability

• Removed an unused, dormant coding-agent backend and stabilized a class of flaky internal tests, so CI is clean.

Install

Linux / macOS: curl -fsSL https://get.vexillon.ai/install.sh | sh · Windows: irm https://get.vexillon.ai/install.ps1 | iex

Or download the signed build for your platform below, verify it against the vexillon.ai release key (below), extract the mira binary onto your PATH, and run mira setup. The web UI is bundled into the binary.

Targets

mira-0.315.1-x86_64-unknown-linux-gnu.tar.gz · -aarch64-unknown-linux-gnu.tar.gz · -x86_64-apple-darwin.tar.gz · -aarch64-apple-darwin.tar.gz · -x86_64-pc-windows-msvc.zip

Verify

Signed with minisign under vexillon.ai:

sha256sum -c SHA256SUMS
minisign -Vm mira-0.315.1-x86_64-unknown-linux-gnu.tar.gz \
  -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO

Trusted comment: signed by vexillon.ai — MIRA 0.315.1.

MIRA v0.313.0 — public beta

MIRA v0.313.0 — Guardian sentinel: faster alarms, reliable delivery, one-click enable

A self-hostable, single-binary personal AI agent that runs on hardware you control. A project of Vexillon. Licensed under AGPL-3.0-or-later.

⚠️ Public beta (pre-release). Recommended build for new installs; supersedes v0.311.0.

This release hardens the out-of-process MIRA-Guardian liveness sentinel — the watchdog that keeps watching even when MIRA itself goes down — based on a live failover test.

Faster, more reliable down-alarms

• Notify-first. The "MIRA is down" alarm now fires immediately when the sentinel declares MIRA down. The optional local-model triage that adds detail runs afterward, as a follow-up — so a slow or unavailable model can never delay the alarm.
• No more silent failures. If an alarm reaches no device, that's now treated as a failure — logged loudly and recorded — instead of looking like a success. The Guardian panel warns when the sentinel is watching but can't reach anyone, and the sentinel checks this at startup.

Enable it with one toggle

• Turning on the liveness sentinel in Settings → Guardian now automatically registers and starts it as its own supervised service (systemd on Linux, launchd on macOS, SCM on Windows). Turning it off stops and unregisters it. No separate mira guardian-install step required (the CLI commands remain for headless setups).

Configurable logging

• The sentinel now shares MIRA's main log file by default, so all logs land together. Prefer a separate file? Set Settings → Guardian → Sentinel log file (guardian.process.log_file).

Reminder

For a down-alarm to reach you when MIRA is offline, register at least one push device — open MIRA in a browser and allow notifications.

Install

Linux / macOS: curl -fsSL https://get.vexillon.ai/install.sh | sh · Windows: irm https://get.vexillon.ai/install.ps1 | iex

Or download the signed build for your platform below, verify it against the vexillon.ai release key (below), extract the mira binary onto your PATH, and run mira setup. The web UI is bundled into the binary.

Targets

mira-0.313.0-x86_64-unknown-linux-gnu.tar.gz · -aarch64-unknown-linux-gnu.tar.gz · -x86_64-apple-darwin.tar.gz · -aarch64-apple-darwin.tar.gz · -x86_64-pc-windows-msvc.zip

Verify

Signed with minisign under vexillon.ai:

sha256sum -c SHA256SUMS
minisign -Vm mira-0.313.0-x86_64-unknown-linux-gnu.tar.gz \
  -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO

Trusted comment: signed by vexillon.ai — MIRA 0.313.0.

MIRA v0.311.0 — public beta

MIRA v0.311.0 — Guardian sentinel installer, now on macOS and Windows

A self-hostable, single-binary personal AI agent that runs on hardware you control. A project of Vexillon. Licensed under AGPL-3.0-or-later.

⚠️ Public beta (pre-release). Recommended build for new installs; supersedes v0.310.0.

The out-of-process MIRA-Guardian liveness sentinel — the watchdog that keeps watching even when MIRA itself crashes, hangs, or restart-loops, and raises a direct alarm to your device — can now be installed as a supervised service on all three desktop platforms, not just Linux.

mira guardian-install registers the sentinel as its own service, deliberately separate from the main MIRA service so it outlives a MIRA crash:

• Linux — a second systemd unit (mira-guardian-watch)
• macOS — a launchd agent (com.mira.guardian-watch)
• Windows — a dedicated SCM service (mira-guardian-watch)

mira guardian-uninstall removes it, and mira uninstall cleans up both the main service and the sentinel.

The sentinel is off by default. Enable and tune it under Settings → Guardian → Liveness sentinel, then run mira guardian-install.

Install

Linux / macOS: curl -fsSL https://get.vexillon.ai/install.sh | sh · Windows: irm https://get.vexillon.ai/install.ps1 | iex

Or download the signed build for your platform below, verify it against the vexillon.ai release key (below), extract the mira binary onto your PATH, and run mira setup. The web UI is bundled into the binary.

Targets

mira-0.311.0-x86_64-unknown-linux-gnu.tar.gz · -aarch64-unknown-linux-gnu.tar.gz · -x86_64-apple-darwin.tar.gz · -aarch64-apple-darwin.tar.gz · -x86_64-pc-windows-msvc.zip

Verify

Signed with minisign under vexillon.ai:

sha256sum -c SHA256SUMS
minisign -Vm mira-0.311.0-x86_64-unknown-linux-gnu.tar.gz \
  -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO

Trusted comment: signed by vexillon.ai — MIRA 0.311.0.

MIRA v0.310.0 — public beta

MIRA v0.310.0 — public beta

The current public-beta build of MIRA — a self-hostable, single-binary personal AI agent that runs on hardware you control. A project of Vexillon. Licensed under AGPL-3.0-or-later.

⚠️ Public beta (pre-release). Recommended build for new installs; supersedes v0.306.0. This is a batch release — it folds in the 0.307–0.310 increments that complete the independent Guardian.

The fully independent, agentic Guardian

This release completes MIRA-Guardian's journey from a co-resident health monitor into a fully independent, out-of-process guardian that keeps watching even when MIRA itself is down. (Still off by default; opt-in, and fail-closed local-only throughout.)

• Agentic triage. The standalone triage can now actively use the Guardian's read-only tools — inspect the full health snapshot and logs, consult the built-in docs, and recall relevant past context — to investigate before it diagnoses, producing sharp, specific next steps.
• Delivered in MIRA's voice when it's up. When MIRA is reachable, the sentinel hands its message to MIRA over a local, shared-secret endpoint, so MIRA delivers it in its own voice across all your channels (Signal, Telegram, email, …). Direct web-push remains the fallback for when MIRA is truly down.
• The sentinel can own the whole watch. An optional setting lets the sentinel take full ownership of health watch and triage — it then also triages problems while MIRA is up, and MIRA's in-process watcher stands down so the two never double-alert.
• Plus a flaky-test fix for CI stability.

Net: an independent local guardian that watches your MIRA's liveness and health, diagnoses issues with a local model using real tools, and reaches you through MIRA when it's up or directly when it's down — private and fail-closed the whole way.

Install

Linux / macOS: curl -fsSL https://get.vexillon.ai/install.sh | sh · Windows: irm https://get.vexillon.ai/install.ps1 | iex

Or download the archive below, extract the mira binary onto your PATH, and run mira setup. The web UI is bundled into the binary.

Targets

mira-0.310.0-x86_64-unknown-linux-gnu.tar.gz · -aarch64-unknown-linux-gnu.tar.gz · -x86_64-apple-darwin.tar.gz · -aarch64-apple-darwin.tar.gz · -x86_64-pc-windows-msvc.zip

Verify

Signed with minisign under vexillon.ai:

sha256sum -c SHA256SUMS
minisign -Vm mira-0.310.0-x86_64-unknown-linux-gnu.tar.gz \
  -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO

Trusted comment: signed by vexillon.ai — MIRA 0.310.0.

MIRA v0.306.0 — public beta

MIRA v0.306.0 — public beta

The current public-beta build of MIRA — a self-hostable, single-binary personal AI agent that runs on hardware you control. A project of Vexillon. Licensed under AGPL-3.0-or-later.

⚠️ Public beta (pre-release). Recommended build for new installs; supersedes v0.305.0.

Real Guardian triage — in the sentinel's own process

When you configure a local model, the out-of-process sentinel now runs a real Guardian triage turn in its own process: a fail-closed, local-only model call that reasons over MIRA's health data and delivers a charter-voiced alert with a concrete next step — all independently of the MIRA that's down.

If no local model is reachable, it falls back to the plain alarm, so an alert never fails to fire.

Install

Linux / macOS: curl -fsSL https://get.vexillon.ai/install.sh | sh · Windows: irm https://get.vexillon.ai/install.ps1 | iex

Or download the archive below, extract the mira binary onto your PATH, and run mira setup. The web UI is bundled into the binary.

Targets

mira-0.306.0-x86_64-unknown-linux-gnu.tar.gz · -aarch64-unknown-linux-gnu.tar.gz · -x86_64-apple-darwin.tar.gz · -aarch64-apple-darwin.tar.gz · -x86_64-pc-windows-msvc.zip

Verify

Signed with minisign under vexillon.ai:

sha256sum -c SHA256SUMS
minisign -Vm mira-0.306.0-x86_64-unknown-linux-gnu.tar.gz \
  -P RWQX8PcjZB3S8i1V9AkZgUKoIy+eZWRzaEHKvBW9U/mkMP9Vp0B8xOaO

Trusted comment: signed by vexillon.ai — MIRA 0.306.0.