Repository navigation
Releases
IRMind v0.6.0
English
Action required: approve settings again in
/settings. If your previous configuration approval stored only a hash, review and sign the settings again before using the saved configuration.
IRMind v0.6.0 adds budgeted source queries for CloudTrail in Athena, Cloud Audit Logs in BigQuery, and supported Azure Log Analytics tables on the Analytics plan. Acquisitions become signed captures for local investigation.
- Incompatible formats. This release uses case marker 5 and export V9. Use the original v0.5.x binary to open, verify or export a v0.5.x case, or re-ingest the verified originals into a new case.
- Opt-in. Source queries are disabled by default. The organization must configure
[source_queries]in its policy, and the analyst must store the provider credentials in IRMind's keystore, adopt the source and allocate a case budget. - Experimental. These three integrations have only been validated against simulators. Real-service validation is pending for Athena, BigQuery and Azure Log Analytics.
- Provider setup. Athena workgroup with a scan cutoff · Cloud Logging sink and BigQuery table layout · Entra app registration and minimum permissions.
Español
Acción necesaria: vuelve a aprobar los ajustes en
/settings. Si la aprobación anterior solo guardaba un hash, revisa y firma los ajustes de nuevo antes de usar la configuración guardada.
IRMind v0.6.0 añade consultas en origen con presupuesto para CloudTrail en Athena, Cloud Audit Logs en BigQuery y las tablas de Azure Log Analytics admitidas en el plan Analytics. Las adquisiciones se conservan como capturas firmadas para el análisis local.
- Formatos incompatibles. Esta versión usa el marcador de caso 5 y la exportación V9. Usa el binario v0.5.x original para abrir, verificar o exportar un caso de v0.5.x, o reingiere los originales verificados en un caso nuevo.
- Activación explícita. Las consultas en origen están desactivadas por defecto. La organización debe configurar
[source_queries]en su política, y el analista debe guardar las credenciales en el keystore de IRMind, adoptar la fuente y asignar presupuesto al caso. - Experimental. Estas tres integraciones solo se han validado con simuladores. Está pendiente la validación con los servicios reales de Athena, BigQuery y Azure Log Analytics.
- Configuración por proveedor. Workgroup de Athena con tope de escaneo · Sink de Cloud Logging y diseño de tablas de BigQuery · Registro de la aplicación en Entra y permisos mínimos.
Code signing policy: https://github.com/ns2switch/irmind/blob/main/CODE_SIGNING.md
IRMind v0.5.7
What's Changed
- Summarize recent releases in both READMEs by @ns2switch in #359
- Check Azure external-sort disk space before ingestion by @ns2switch in #361
- Reduce external sort passes with bounded 64-way merges by @ns2switch in #362
Full Changelog: v0.5.6...v0.5.7
Code signing policy: https://github.com/ns2switch/irmind/blob/main/CODE_SIGNING.md
IRMind v0.5.6
What's Changed
- Speed up external sorting for disordered Azure flow logs by @ns2switch in #358
Full Changelog: v0.5.5...v0.5.6
Code signing policy: https://github.com/ns2switch/irmind/blob/main/CODE_SIGNING.md
IRMind v0.5.5
What's Changed
- Describe current signing and remove sponsorship attribution by @ns2switch in #327
- Design bounded source-side queries for v0.6 by @ns2switch in #326
- Design source-query ports, signed acquisition and export V9 by @ns2switch in #328
- Add bounded M365 and Okta source modules by @ns2switch in #331
- Add GuardDuty source findings and bounded envelope decoding by @ns2switch in #332
- Add experimental AWS, Azure and GCP network flow modules by @ns2switch in #333
- Add bounded email export decoding and identified mappings by @ns2switch in #334
- Add Defender export decoding and declarative event mappings by @ns2switch in #335
- Add validated native FDR process identities by @ns2switch in #337
- Decode and map bounded native Falcon FDR exports by @ns2switch in #338
- Add experimental macOS Unified Log and eslogger export modules by @ns2switch in #341
- Report Windows EVTX witness exits and receipt rejection reasons by @ns2switch in #345
- Scope expanded mapping registries to bounded signed profiles by @ns2switch in #348
- Add experimental Google Workspace audit export modules by @ns2switch in #346
- Decode and map GitHub audit log exports by @ns2switch in #347
- Decode and map Route 53 Resolver query logs by @ns2switch in #349
- Decode Windows DNS debug logs and align EVTX DNS mappings by @ns2switch in #350
- Decode IIS W3C access logs with declarative mapping by @ns2switch in #351
- Decode AWS ALB and Classic Load Balancer access logs by @ns2switch in #352
- Record Windows EVTX startup diagnostics in custody tests by @ns2switch in #357
- Decode original Amazon S3 server access logs by @ns2switch in #353
- Bundle the v0.5.5 core skills from the upstream mirror by @ns2switch in #354
- Prepare IRMind v0.5.5 by @ns2switch in #356
Full Changelog: v0.5.4...v0.5.5
Code signing policy: https://github.com/ns2switch/irmind/blob/main/CODE_SIGNING.md
IRMind v0.5.4
IRMind v0.5.4 — Verified Homebrew installation and the community skill catalog
Format impact: No case, artifact, seal, export or definition format changes.
The catalog's production Ed25519 public key is now compiled into the accepted
key list. Synthetic fixture keys remain excluded from release trust.
- Install the main executable and EVTX helper from the Homebrew tap on macOS
Apple Silicon/Intel and Linux x86_64/arm64. Release dispatch authenticates
checksums and runs native installation, audit, demo and uninstall checks. - Verify the independently published community catalog with its pinned production
key; installation remains explicit, previewed and admitted by an analyst signature.
Community contributions, synthetic episodes and review guidance live in
ns2switch/irmind-skills, separately from the core skills bundled here. - Publish the project code signing policy and release-note link. Windows
Authenticode and Apple notarization are not activated by this release.
Evaluation scope
Pre-1.0 evaluation software, not validated for real incidents. An analyst has recorded a partial multi-turn investigation on IRMind 0.5.3 with DeepSeek and locally ingested public CloudTrail data, including a skill and confirmed findings. Elastic with TLS/private CA, hash citations, /verify and export verification remain pending analyst observations; see the validation record.
Keep the exact executable and EVTX helper that created a case before upgrading. Each pre-1.0 minor version may change case, export and definition formats without migration. This patch does not change those formats.
Installation and verification
Follow the verified installation guide. On macOS, the Homebrew tap installs the binary and helper with pinned checksums; its version update follows native checks. Native Apple notarization and Windows Authenticode are not active. Sigstore bundles, SHA256SUMS, provenance and both SBOMs are attached for independent verification.
The catalog production public key is compiled into this release. Catalog releases are published independently in irmind-skills; package installation remains explicit and analyst-signed.
Code signing policy: https://github.com/ns2switch/irmind/blob/main/CODE_SIGNING.md
IRMind v0.5.3
IRMind v0.5.3 — Prompt-cache accounting and interrupted-plan recovery
Format impact: Optional cache metadata extends signed usage receipts and
optional cache prices extend signed configuration. Missing fields retain their
existing canonical bytes. Case, artifact, seal and export versions are unchanged.
Retain the creating executable for full reconstruction of its signed native decoder/mapping profiles.
-
Replace unmaintained calendar/duration dependencies with maintained date
primitives and narrowly reviewed Arrow/Parquet/storage source patches. Current
canonical and physical-format golden bytes are unchanged; advisory enforcement,
explicit credentials, TLS, bounded reads and EVTX isolation remain enabled. -
Publish native macOS Intel binaries alongside Apple Silicon, Linux x86_64,
Linux aarch64 and Windows. Verify clean installers even when optional native
signing is disabled. Sigstore, provenance and SBOM verification remain mandatory. -
Refresh the complete offline presentation with the actual v0.5.3 executable.
Native Developer ID/notarization and Authenticode remain unavailable. -
Expand the controlled Elasticsearch/TLS rehearsal guidance and the optional
HuntEval pilot runner. A local-model rehearsal does not replace analyst
validation of investigation quality or a connector in a real test environment. -
Recover an interrupted subagent plan explicitly with
/subagents recover,
after confirmation. Nonblocking interprocess activity/recovery locks protect
live previews and children, remain held through reconciliation and prevent
concurrent deletion. Recovery preserves committed work and uncertain usage
without replaying provider requests or tool effects. Signed formats are unchanged. -
Account for provider-reported prompt-cache reads and writes separately in
/budgetand the configurable status bar. Inclusive input still consumes the
full token budget; estimates require every applicable price. Native Anthropic
requests mark the trusted system/tool prefix; OpenAI and DeepSeek cache
automatically. Compatible/local endpoints receive no new cache controls.
Optional cache fields extend signed usage receipts and price configuration;
absent fields preserve existing bytes. Case, seal and export versions are unchanged. -
Show an actionable notice when a provider completes without a visible answer,
while retaining its signed empty reply and reported usage. -
Rebuild mapped evidence through its signed decoder/mapping profile in both CLI
and TUI; connector decoders do not need automatic file-format detectors. -
Verify private originals retained by the case at their signed location instead
of requiring a second external duplicate of acquired connector responses. The
same signed source location is used when including originals in an export.
Missing, altered or unavailable retained originals still fail verification. -
Confirmed Ctrl+C exits the application from an open case; canceling its modal
leaves the conversation unchanged. Evidence-picker help matches the actual keys.
Evaluation limits
IRMind remains pre-1.0. Validate a configured model and connector against your own controlled test environment before relying on their investigative results. The local synthetic rehearsal does not establish investigation quality. Native macOS Developer ID/notarization and Windows Authenticode are not active; verify the Sigstore signatures, GitHub provenance and SHA-256 checksums. The external skills catalog is not available in this release; bundled skills remain available.
Full changelog · Verified installation and platforms · Controlled validation
Code signing policy: https://github.com/ns2switch/irmind/blob/main/CODE_SIGNING.md
IRMind v0.5.2
IRMind v0.5.2
Format impact: Case, artifact, seal and export versions are unchanged. Adds
an optional signed status-bar configuration and the bounded case-deleted-v1
global operation; existing configuration and global-envelope golden bytes are
unchanged. Pre-1.0 minor versions may change formats without migration: retain
the creating IRMind version or re-ingest verified originals into a new case.
- Delete an unsealed case with
/case delete, after reviewing its inventory and
typing the exact title or identifier. An Analyst-signed global tombstone and
durable intent commit together before cleanup. Interrupted cleanup resumes from
authenticated receipts;/cases deletedretains the audit. Sealed cases and
cases open in another process cannot be deleted. External originals and evidence
shared with another case are preserved. Deletion is logical, not secure erasure. - Customize the status bar in Settings → Preferences with a bounded list of
segments, priorities, theme colors and separators. Includes live preview,
reset and minimal/investigation/cost presets, including 80-column terminals and
NO_COLOR. Mandatory scope and budget warnings remain visible. Source counts
come from signed manifests even when originals are not copied. - Save provider, URL, model, native output maximum and parameter requirement as
one validated, Analyst-signed model configuration. Connection testing uses the
complete unsaved draft; known cross-provider combinations are rejected. - Link bundled skills and contribution guidance to the separate skills repository,
distinguishing release-bundled core methods from the optional catalog. - Avoid repeated keystore credential enumeration when refreshing the case footer,
and ignore key-release events when returning from the keystore wizard to home.
Native Developer ID/notarization and Authenticode are not active. The binaries
carry Sigstore signatures, provenance attestations and an SBOM; operating systems
may still display native-signing warnings. Real-service coverage remains limited;
use the documented controlled-lab validation before relying on a new environment.
This is a pre-1.0 evaluation release. No provider or connector is enabled by default. The local synthetic demo needs no model setup or commercial key.
Verify the signed checksum list, binary bundles and provenance before running downloaded files. Installation and verification instructions: English · Español. The packages include the isolated EVTX helper and separate SBOMs.
Known limitation: mixed cases containing uncopied local files and retained connector responses can fail original verification, graph reconstruction or export with originals in this build. Corrections are tracked in #309 and are not included in v0.5.2. Preserve the creating executable and verified originals.
IRMind v0.5.1
IRMind v0.5.1 — Complete model responses and evaluation readiness
Format impact: Case, artifact, skill and export versions are unchanged. Model
replies in subagent branches may now use the available 1 MiB frame capacity;
other branch payload limits are unchanged. Model-token settlement accepts signed,
verified usage above an already admitted reservation without creating capacity.
- Use native model output maxima instead of application token caps, with signed model capability settings for custom/local endpoints. Enforce case budgets between requests and retain completed responses and actual usage when they overdraw the balance.
- Continue a provider-limited answer once without tools when budget permits, keeping one continuous answer. Retain full in-frame child replies separately from their 16 KiB parent summaries; label security-boundary prefixes explicitly.
- Test model connections with the current unsaved provider, URL, model, key and output profile; probing does not save configuration or credentials.
- Show case help in a scrollable view and resume the conversation at its end after the demo presentation.
- Refresh the verified-install quick start and synthetic v0.5.0 images, and add clean-release PTY checks for Linux, macOS and Windows.
- Add opt-in real-service validation and multi-model HuntEval development helpers. Prepare disabled tag-only Developer ID signing and notarization; native signing remains unavailable until provisioned and validated.
Known limitations
- Pre-1.0 evaluation with synthetic data. A real DeepSeek greeting was accepted by the analyst; the remaining investigation procedure and a real connector validation are still pending. This is not a production incident-response validation.
- Existing 1,000,000-byte guard, 2,000,000-byte stream and 1 MiB signed-frame limits remain. A retained prefix at a security boundary is labelled incomplete.
- Native Developer ID/notarization and Authenticode remain unavailable; use the Sigstore bundles, exact workflow identity, GitHub attestations and SHA256SUMS.
- No production catalog public key is compiled yet. Core skills remain available; catalog publication is separate.
- Context/caching changes are design-only in PR #302.
See the configuration guide, real validation guide and verified installation instructions.
IRMind v0.5.0 — Bounded parallel investigations
Format impact: New cases use CaseCreated version 4 with an initial signed
allocation. Export V8 replaces V7 and always includes catalog-proof and subagent
branch counts, including zero counts. Older cases/exports require their creating
IRMind version or re-ingestion of verified originals; there is no automatic migration.
- Delegate bounded investigations to six included subagent types with progressively
loaded skills, source-restricted queries and shared case budgets. The TUI shows
plans before start, requires signed confirmation above 25% of remaining tokens
or calls, and offers signed confirmation for every plan as a preference. - Stop before an object transfer that cannot fit its task byte allocation, preserving a partial result with the remaining budget.
- Keep a partial, verifiable task result when a provider repeats a tool-call identifier, without repeating its effect.
- Follow activity and partial coverage with
/subagents; cancel individual tasks
without discarding completed work. Child approval requests return to the owning
live task, and siblings continue while an Analyst decision is pending. - Verify and export Agent-signed child branches, canonical checkpoints, task-owned
remote acquisitions and deterministic results. Restart recovery closes
interrupted work without retrying remote effects. Children cannot delegate,
approve requests, change settings or expand their assigned sources or scope. - Honor validated host, cluster, category, type, outcome and entity-role filters
in local model-tool queries. - Use SkillInstallationV2 and SkillSnapshotV2 for current methodology admissions.
Catalog provenance requires a signed index from a compiled publisher key and
an independent, immutable proof in every referencing case. Export V8 carries
exactly the proofs needed for offline verification. No production catalog key
is configured yet, so catalog publication and downloading remain unavailable. - Establish the shared case/export foundation for catalog admissions and
subagent branches, retaining current seal bytes and frame limits.
Evaluation limits
- Preliminary release for synthetic-data evaluation, not real incidents or sensitive evidence. The accepted real-model H5 observation covers a DeepSeek greeting; the remaining procedure details and real-connector validation are still pending.
- Six included subagent types only. Community types and monetary budget limits are deferred; costs are estimates or unavailable.
- Instance-scoped SIEM queries are unavailable to children. Remote adapters without a completeness contract retain explicit partial coverage.
- No production catalog publisher key is configured. Catalog publication/downloads remain unavailable; bundled and local skill installation work without catalog access.
Verify the download
The release includes Linux x86_64/aarch64, macOS aarch64 and Windows x86_64 binaries, the isolated EVTX helper, installers, SHA256SUMS, exact-tag Sigstore bundles, GitHub provenance and two SPDX SBOMs. Follow the verification guide before executing a download. The expected signing identity is https://github.com/ns2switch/irmind/.github/workflows/ci.yml@refs/tags/v0.5.0 and the issuer is https://token.actions.githubusercontent.com.
Parallel investigations · Guía en español · Case version policy
IRMind v0.4.2
IRMind v0.4.2 — Case budgets and Kubernetes investigations
Format impact: current case contracts change. The expanded common-event
vocabulary changes the mapped target-contract digest. Cluster-aware sources add
explicit acquisition context and optional physical metadata; new case allocations
are signed in creation/change receipts. Export V7 framing and seal challenges
are unchanged, but their current profile/envelope vectors reflect the new
contract. Retain the creating executable to open, verify or export an older
case, or re-ingest verified originals into a new case. No automatic migration
is provided.
All modules and methods remain experimental and tested with synthetic data.
H5 was accepted by the analyst after a real DeepSeek greeting; the remaining
manual procedure details and validation of a real connector remain pending.
Kubernetes investigation methods and verified originals
- Include
k8s-cluster-compromiseandk8s-workload-compromise, bringing the core
catalog to 15 skills. Both cover source collection, typed local queries,
original-field review, coverage gaps and conditional findings. Their scripted
synthetic episodes confirm findings and verify, seal and independently verify
exported cases; expected answers remain evaluator-only. - Add explicit
raw_records view="original"for bounded, guarded source facts.
Local source roots require analyst-origin authorization. Reads verify source
bytes, the frozen decoder/context and the signed decoded-record commitment;
indexed reads remain the default. No network, decoded cache or spool is used. - Add the separate Falco Kubernetes-audit mapping and explicit privilege facts.
Kubernetes protocol-upgrade responses remain unknown outcomes, rather than
evidence that a command ran successfully. - Cover native audit logs and managed EKS, GKE and AKS exports. No live Kubernetes
API connector, remote model default or implicit cloud-identity link is added.
Fixed
- Align the isolated EVTX fuzz workspace with the current helper version and
check resolution of all four locked workspace graphs during CI. Keep the
existing decoder and protocol ASAN campaign budgets.
Managed Kubernetes export transports
- Decode EKS CloudWatch GetLogEvents and AKS kube-audit/kube-audit-admin
resource-log envelopes, preserving original transport fields and the decoded
audit Event with real outer-row provenance. - Include separate hash-pinned EKS/AKS tables and explicit cluster confirmation.
Reject contradictory source claims and request confirmation for new clusters.
Detect GCP/GKE LogEntry arrays produced by gcloud as well as NDJSON. - Format impact: additional source-format/mapping identities; native Event
decoding, signed envelopes, physical schemas, seals and exports do not change.
Microsoft Entra source coverage
- Add experimental, default native SignInLogs and AuditLogs decoding and
identified mappings for Azure Monitor resource logs and Log Analytics object
rows, including non-interactive and service-principal sign-ins. - Preserve nanosecond time and original dynamic/MFA/Conditional Access fields;
map explicit outcome and tenant-scoped actor/target roles. Update the Azure
skill and its complete scripted episode to use native semantic queries. - Format impact: new parser/mapping identities, with no envelope, schema,
seal or export framing change.
Network interpretation
- Classify native Zeek and Suricata records with typed endpoint roles, protocol
fields, explicit status outcomes and unconfirmed source alerts. Update the
network skill's synthetic episode to query category, outcome and role. - Network mapping identities and their current logical golden projections change;
source decoders, signed envelopes, physical schemas and export format do not.
GCP Cloud Audit and GKE interpretation
- Add a default, experimental Cloud Audit JSON/NDJSON source decoder and bundled
service/status mappings, sharing explicit cluster confirmation with GKE.
Preserve nanosecond time and original fields; distinguish reported caller,
target resource and source/destination endpoints. Redacted IPs remain unknown. - Keep missing RPC status unknown and classify project-policy operations only
from documented exact method names. Validate GKE namespace/resource paths and
keep zonal locations distinct from regions. Unknown services retain coverage gaps. - Update the GCP service-account skill and its scripted episode to use native
IAM/outcome/role filters, with independent evaluator-only HuntEval answers. - Format impact: no additional envelope, export framing or seal change.
Kubernetes and Falco source modules
- Add separate experimental native Kubernetes audit and Falco JSON decoders, selected by the bounded detection registry. Original fields, nanosecond timestamps and actual offsets are retained; mappings remain independent.
- Ask for required cluster acquisition context in the TUI, with explicit per-file or batch confirmation. CLI and
/ingest --clusteruse the same typed, signed context. No identity is inferred from the machine or path, and no live API access is added. - Include hash-pinned Kubernetes and Falco table mappings: explicit API outcomes, actor/impersonation roles, namespaced target resources, and unconfirmed source alerts. Cluster search projects validated fields instead of whole request/response bodies.
- Include synthetic fixtures, detection-matrix coverage, malformed/oversized-record regressions, cluster-confirmation keyboard tests and signed ingest/rebuild/seal/export coverage.
- Format impact: no additional format change beyond the cluster/common vocabulary extensions described below.
Cluster-qualified common events
- Extend the current common event vocabulary with bounded orchestrator,
container, group and privilege fields and cluster/namespace-qualified entities.
Raw producer cluster IDs remain separate from acquisition identity. - Add an exact
clusterquery filter, signed partition pruning and row filtering
inside overflow buckets. Hash verification and all query limits remain intact. - Cluster artifacts carry an explicit optional column/metadata extension;
non-cluster Parquet files retain their physical bytes. Unmapped cluster sources
retain originals and commitments without projecting arbitrary request bodies. - Format impact: the expanded common vocabulary changes the mapped target-contract
digest and current profile/envelope/export vectors. V7 framing and seal
challenges are unchanged. Use the creating version for another contract, or
re-ingest verified originals into a new case; no migration is introduced.
Kubernetes ingestion foundation
- Bind explicit, bounded cluster acquisition context to source manifests,
decoded-record commitments and reconstruction. Cluster claims cannot grant
permissions or silently replace analyst-confirmed identities. - Add deterministic cluster partitions with 64 ordinary values by default,
a maximum of 1,024 and signed overflow buckets. Raw cluster identifiers never
become directory names. - Format impact: current source manifests and record-set commitments gain an
optional cluster-context extension; records carrying cluster context include
it in their canonical hash. Sources without context retain their bytes.
Case allocations
- Add signed defaults for new cases in
/settings limitsand signed per-case
token, tool-call and reserved cloud-byte allocations in/budget. Organization
ceilings remain mandatory; increases need explicit Analyst confirmation. - Use 2,000,000 tokens and 1,000 calls by default, retaining all fixed maxima and
existing cloud-transfer controls. Existing policy files are not rewritten. - Show consumption and 80% warnings, and retain completed work with a tool-free
or local summary when an allocation is exhausted. Missing provider usage is
labeled unavailable; costs are shown only when supported by usage/pricing. - Format impact: case creation records include the initial allocation and later
changes useCaseBudgetChangedreceipts. Existing signed-entry and export V7
framing are unchanged. Cases without a recorded allocation require explicit signed confirmation in
/budgetbefore new model/tool work; recorded consumption is retained.
Download each binary with SHA256SUMS and its Sigstore bundle. Verify the exact tag/workflow identity and GitHub build provenance before installation; see verification instructions. Known co-observation/query performance follow-ups remain in issue #37. macOS notarization and Windows Authenticode are not provided; operating-system warnings may appear.