Docs / Changelog

What changed, in one clean timeline.

A short release history that keeps the focus on what shipped instead of repeating the same product pitch on every entry.

Release 28

v1.14.1

July 2026patch

Added

Context Control V1 now reconciles declarative ProjectContext manifests with tenant-scoped hosted documents through reviewed diff/apply plans
Companion 3.5.1 emits plan-bound Decision Requests and detailed hosted apply/reindex receipts

Security

Hosted apply requires EDITOR API-key access, atomically rechecks remote hashes, allows only add/update, never deletes unmanaged documents, and blocks managed authority promotions
Local Context Control V0 now enforces its explicit review acknowledgement before apply

Release 27

v1.14.0

July 2026minor

Added

Owner operating profiles add an explicit, reviewable USER-scoped home for communication, decision, autonomy, risk, evidence, product-principle, and working preferences
Session bootstrap now reserves bounded project/client and owner profile context before active decisions, other durable memory, and daily carryover

Improved

Project/client profiles now use canonical Memory V2 storage while preserving additional tenant profiles and scoped legacy read compatibility
Generic compaction protects project/client and owner profile categories from ordinary archival, conflict normalization, and deduplication
Public MCP documentation now distinguishes the 128 canonical tools, lean 10-tool default manifest, and 251-entry wire registry

Release 26

v1.13.6

June 2026patch

Added

Engineering Lead Execution Receipts V1 now adds explicit handoff, claim, approval, proof, outcome, and Project Brain update gates to Project Health engineeringLeadPlan exports
snipara-companion 2.3.0 emits and imports executionReceipts, and surfaces unknown future receipt enum values with visible fail-closed reason codes
snipara-orchestrator 0.1.11 can dry-run lead-plan coordination and check receipt evidence with agents coordinate --lead-plan-file and agents check-receipt

Improved

Project Health, Companion, and orchestrator now share one receipt ledger while keeping workersSpawned at 0 until explicit execution evidence exists
Missing approval, proof, outcome, and Brain-update receipts now become observable diagnostics instead of implicit readiness assumptions

Release 25

v1.13.5

June 2026patch

Added

Engineering Lead Contract V1 now adds a stable contractVersion, supervised work packages, supervision status, review requirements, replan triggers, and evidence refs to Project Health engineeringLeadPlan exports
snipara-companion 2.2.0 can import plans with --from-plan and reconcile them with --reconcile before treating an engineering lead plan as current
snipara-orchestrator 0.1.10 can route from a selected Engineering Lead Contract V1 work package in dry-run mode with --lead-plan-file

Improved

Unknown future Engineering Lead Contract enum values now fail closed with visible reason codes instead of silently downgrading imported cockpit plans
Project Health and Companion now expose supervision and replan signals without autonomously launching workers

Release 24

v1.13.4

June 2026patch

Added

Project Health now exposes repoIntelligence with a compact risk level, headline, explanation, factors, evidence, and recommended action derived from existing outcome, decision, verification, reason-code, and advisor influence signals

Improved

Outcome attribution closure now persists linked, missed, and visible-only advisor receipt outcomes so Project Health can explain why advice-to-result loops closed
Reason-code calibration now classifies each bin as reliable, weak, contradicted, under-sampled, or unevaluated, with class counts available to API clients

Release 23

v1.13.3

June 2026patch

Improved

Project Intelligence recommendations now use evaluated reason-code calibration to rank competing explanations, rerank brief candidates, calibrate Project Health guidance, and cite concrete decision/source refs when enough outcome evidence exists
Public docs now distinguish shipped outcome-ranked Brief/Judgment behavior, gated posterior-based memory recall reranking, and opt-in context/chunk outcome reranking
Gated memory recall reranking now uses sample-aware OutcomeStats posterior evidence instead of raw Memory.evidenceScore, while context/chunk reranking now prefers persisted OutcomeEvidenceStats with live fallback, bounded boosts, and a top-relevance floor

Release 22

v1.13.2

June 2026patch

Improved

Agent docs now make snipara-companion code impact the primary shell-agent impact gate, with local overlay for dirty or ahead worktrees and hosted MCP as fallback or canonical post-reindex check
Generated agent templates now include companion memory invalidate and memory supersede commands for correcting stale durable memory without deleting history
Project Intelligence docs now clarify that advisory calibration only evaluates direct or scoped outcome attribution; temporal or ambiguous links remain unevaluated evidence

Release 21

v1.13.1

June 2026patch

Added

Project Intelligence docs now explain the five-question model for agent work: what changed, why, impact, next step, and whether it is safe to proceed
Public blog now includes Project Intelligence, Not Just Project Memory as a positioning article separate from package release notes

Improved

Homepage messaging now separates shipped Project Intelligence surfaces from emerging ones instead of overclaiming autonomous project judgment
snipara-companion 1.4.5 release notes now document conservative Team Sync work-item cleanup after completed workflow commits

Release 20

v1.13.0

June 2026minor

Added

snipara-companion local code overlays now expose code status, sync, upload, hooks, promote, local graph queries, and local MCP server workflows for pre-push agent context
Companion-scoped hosted overlay storage now supports upload, status, get, and retire for non-canonical local code manifests with TTL expiry and repo root redaction
Managed Git hooks can refresh local overlays after commits and record push-time promotion state before hosted CODE reindex catches up

Improved

Agent and code graph docs now separate canonical hosted snipara_code_* answers from local-only overlays for dirty working trees and local commits
Companion workflows now make local overlay freshness visible without silently mixing local state into hosted project truth

Fixed

Hosted overlay upload now handles expiry timestamps correctly during Postgres insert, verified with upload, status, get, and retire smoke tests

Release 19

v1.12.0

May 2026minor

Added

Issue-aware Jira and Linear continuity now enrich Start Work, Team Sync, What Changed, and PR Answer Packs through a shared provider-neutral contract
Operational context graph foundation now persists repository, runtime, coordination, and external work-item slices in Postgres and adds additive graph summaries to What Changed
Multi-agent coordination now ships swarm discovery, capability labels, heartbeat and liveness summaries, stronger claim recovery, and hardened hierarchical task flows
Governance and lineage surfaces now expose authority snapshots, derived claims, and rollout verification for evidence-heavy continuity paths

Improved

Public agent docs now distinguish shipped swarm and htask coordination from older generic future-state copy
Release notes now separate issue-aware context, operational graph, coordination, and governance boundaries more explicitly

Release 18

v1.11.6

May 2026minor

Added

Mistral Le Chat, Mistral Vibe, and LangChain ChatMistralAI tool-calling setup are now first-class automation targets
create-snipara and snipara-companion now generate Mistral MCP-first instructions, Vibe config, Le Chat connector references, and LangChain bindTools examples
Public integration docs now include a dedicated Mistral guide and downloadable Mistral agent templates

Improved

Codex setup now standardizes on SNIPARA_API_KEY, validates the real hosted MCP tools/list surface, and tells users to restart Codex after config changes
Hosted MCP tool contracts now advertise only snipara_* tool names while preserving internal dispatch compatibility

Release 17

v1.11.5

May 2026patch

Added

Snipara best practices now document hosted MCP-first agent workflow, LITE/FULL mode selection, code-impact gates, Sandbox usage, and durable memory persistence
Copy-ready AGENTS.md guidance now tells agents to validate hosted MCP with a tool-oriented call at thread start instead of treating empty resource lists as an outage
create-snipara generated AGENTS.md templates now carry the same hosted MCP startup and memory hygiene guidance as the public template

Improved

AI agent docs now align Codex setup around direct streamable HTTP Hosted MCP configuration and environment-based API keys
Repository AGENTS.md now separates document/context health from indexed code graph freshness and keeps local file reads/tests as the working-tree source of truth

Release 16

v1.11.4

May 2026patch

Added

Public Agent Code Impact launch article with measured hosted MCP capacity signals and usage limits
Billing plan copy now positions snipara_code_symbol_card and snipara_code_impact as paid Context Pro+ features
AI agent templates now tell Codex, Claude Code, and Cursor when to use symbol cards and impact planning

Improved

AI agent integration docs now include a dedicated Agent code context section with Free versus Pro+ behavior
MCP setup UI now reflects the generated hosted contract and lists the Pro+ code impact tools

Release 15

v1.11.3

May 2026minor

Added

Agent-ready code context tools for paid Context plans: snipara_code_symbol_card and snipara_code_impact
snipara_code_impact now accepts changed file sets and returns risk scoring, recommended actions, coverage gaps, and evidence counts
Hosted MCP tools/list now annotates premium code context tools with required plan and availability metadata

Improved

AI agent integration docs now describe when to use symbol cards and impact analysis before risky code changes
MCP code graph reference docs now distinguish basic structural graph tools from Pro+ agent-ready code impact planning

Release 14

v1.11.2

May 2026patch

Added

GitHub-native memory compiler contract and fixture-backed importer groundwork for reviewed agent memory
Public docs now describe GitHub as an evidence layer for future reviewed Memory V2 promotion

Fixed

Updated public MCP tool count copy to the current 121-tool hosted contract
Raised the Python multipart security floor to the patched release used by Dependabot alerts

Security

Resolved Dependabot GHSA-pp6c-gr5w-3c5g alerts by requiring python-multipart 0.0.27 or newer in Python lockfiles

Release 13

v1.11.1

May 2026patch

Improved

Hosted MCP embedding inference now runs through a dedicated internal service instead of loading the large model in every MCP worker
Embedding inference can now be moved to a dedicated VPS with remote readiness checks and firewall-restricted access
Dedicated embedding nodes now keep the model app non-root behind a minimal proxy on the restricted service port
Production worker counts were retuned after live benchmarking: three API workers and two embedding workers
Backend worker count was reduced while keeping a single shared embedding process for lower memory pressure
Backend readiness now verifies the internal embedding service before accepting semantic MCP traffic

Release 12

v1.11.0

May 2026minor

Added

Memory hygiene diagnostics for hosted MCP agent memory, including active anomaly counts and bounded review samples
Reviewable cleanup candidate tools for memory noise, stale rows, duplicate groups, category anomalies, and pending memory review items
Duplicate-only memory candidate review for teams that want focused supersession workflows
Strict hosted MCP memory hygiene smoke validation for deployment checks

Improved

MCP Tools Reference updated to the current 121-tool hosted contract
Agent Memory docs now cover diagnose-first cleanup flows before invalidation, supersession, or compaction

Fixed

Memory invalidation now also marks matching legacy AgentMemory rows inactive so old memories do not keep appearing in health scans
Historical active superseded-category memory anomalies were cleaned from the hosted Snipara project

Release 11

v1.10.0

April 2026minor

Added

Code Context dashboard model for separating reusable engineering rules from project-scoped truth
Project Context collections for reusable source-backed documentation and reference diagrams
Code Context for workspace-wide engineering rules and technical defaults
Project scope filters for client, internal, code, and research projects
MCP tools for reusable uploads and project workspace creation
GitHub Auto-Sync now creates or repairs repository push webhooks when Auto-Sync is enabled

Improved

Dashboard navigation now keeps Code Context for AI context and Settings for workspace administration, members, and API keys
Code Context now uses a quieter tabbed workbench with one primary action per active context area instead of duplicated shortcut cards
Public docs now explain Code Context, Settings, and project scopes
snipara-mcp and snipara-companion now forward single-file upload metadata for project context health
Homepage messaging now describes developer context versus project truth without adding connector complexity
GitHub Sync settings now show explicit webhook health states and manual fallback details when GitHub denies hook management

Release 10

v1.9.0

April 2026minor

Added

Binary parser lane now supports PDF, DOCX, PPTX, SVG, and VSDX files in GitHub sync
Deterministic extraction for page text, document structure, slide content, speaker notes, and Visio topology
GitHub sync now preserves raw binary payloads before indexing instead of coercing them into text

Improved

GitHub Auto-Sync now presents docs, code graph, and binary lanes with clearer file-type guidance
Binary parser reference and public docs now describe the real product scope instead of the SVG-only preview

Fixed

Project sync can now safely ingest supported binary files without corrupting bytes during fetch

Release 9

v1.8.0

April 2026minor

Added

MCP reindex tools: snipara_reindex, snipara_index_health, and snipara_index_recommendations
Code graph tools for structural questions: snipara_code_callers, snipara_code_imports, snipara_code_neighbors, and snipara_code_shortest_path
snipara-companion workflow routing for shared context, code graph, and auto-mode shortcuts
Project-scoped shared context collections so Snipara and Vutler can keep guidance separate

Improved

snipara_context_query stays doc-first while surfacing code-graph hints when relevant
Shared context responses now distinguish team contexts from linked project collections
Public docs and integration guidance now favor slug-first MCP setup paths

Fixed

Shared collection linking now uses a Prisma-compatible code path on the MCP backend
Low coverage now produces reindex recommendations instead of silent drift

Release 8

v1.7.0

April 2026minor

Added

Memory V2 lifecycle tools: snipara_memory_invalidate, snipara_memory_attach_source, snipara_memory_supersede, and snipara_memory_verify
Source-linked memory evidence for facts, decisions, and working context
Evidence verification and stale-memory detection for agent memory workflows
Bulk memory sync support for Memory V2 migration and lifecycle operations

Improved

Public MCP Tools Reference updated to 50 tools with Memory V2 coverage
Agent Memory documentation now covers evidence, freshness, invalidation, and supersession workflows
Public infrastructure pages now reflect production on Infomaniak VPS with Vaultbrix PostgreSQL

Fixed

Memory lifecycle follow-up calls after a fresh store now resolve correctly during dual-write
Memory verification schema alignment for MCP clients

Release 7

v1.6.0

February 2026major

Added

Snipara unified Python SDK v0.1.0: .snipara.toml config, async/sync clients, CLI, file watcher, auto-feedback loop (pip install snipara)
Snipara Sandbox v2.0.0: autonomous agent runner, sub-LLM orchestration, WebAssembly REPL
4 new MCP orchestration tools (43 total): snipara_load_document, snipara_load_project, snipara_orchestrate, snipara_repl_context
Snipara Sandbox MCP Server: 7 tools for Claude Code (execute_python, REPL context, agent run/status/cancel)
Feb 2026 benchmarks: 9.5/10 coding quality, 72% token reduction, 62% cost savings, 0.86 CBS score
Structured outputs with JSON schema constraints in Snipara Sandbox
Multi-modal support (images, audio) in Snipara Sandbox
Streaming support for real-time token output

Improved

MCP tool count: 39 → 43 with new Orchestration category
Benchmark section updated with Phase 8 real data
Snipara Sandbox documentation rewritten for v2.0.0
Landing page integrations section now includes Python SDK

Release 6

v1.5.0

January 2026minor

Added

ChatGPT integration via GPT Actions with OpenAPI schema generation
OAuth 2.0 provider for ChatGPT Apps authentication
Codex CLI integration support
Gemini CLI integration support
VS Code extension v1.2.0 with Snipara Sandbox bridge (Docker/local execution, logs, visualizer)
15 GitHub Copilot Language Model Tools in VS Code extension
43 VS Code commands (up from 34)

Improved

Team collection search now works in project shared context
Shared context search, discovery, and project UI/UX improvements
Mixpanel autocapture and identity reset hardening
MCP config generator updated to reflect 43 total tools

Fixed

Session context overflow in shared context queries
Null-safety crashes on agents, billing, and analytics dashboard pages
Agent memory and group memory documentation aligned with actual MCP API parameters
Multi-agent documentation corrected (roles, ghost tool references removed)
API reference page broken links now use anchor navigation
Dashboard accessibility: keyboard navigation for interactive cards and dropdowns

Release 5

v1.4.0

January 2026minor

Added

39 MCP tools now documented (up from 21)
14 Snipara Agents tools added to MCP reference with deep linking
Document sync tools: snipara_upload_document, snipara_sync_documents, snipara_settings
Agent Memory tools (4): snipara_remember, snipara_recall, snipara_memories, snipara_forget
Multi-Agent Swarm tools (10): swarm creation, resource locking, shared state, task queues
Snipara Agents section in MCP Tools Reference with product cards

Improved

MCP Tools Reference now shows tool count (39 tools)
Feature availability table includes Agents subscription info
Deep linking support for all agent tool documentation

Release 4

v1.3.0

January 2026minor

Added

Interactive demo page showing the READ-EVAL-PRINT-ANSWER pipeline
Automation hooks for context preservation during compaction
Auto-inject context to Snipara cloud on compaction events
Support for project slugs in API routes

Improved

MCP server now auto-injects usage instructions into query responses
Better error messages for API key validation
Updated documentation with hook configuration examples

Fixed

Session context persistence across compaction events
API key validation for project-specific routes

Release 3

v1.2.0

January 2026minor

Added

Shared context tools: snipara_shared_context, snipara_list_templates, snipara_get_template
Shared context collections for team-wide coding standards
Prompt templates feature for reusable prompts
GitHub account linking for email users

Improved

Updated MCP dependency to version 1.25.0
Better token budget allocation by document category
Redis caching layer for faster query responses

Release 2

v1.1.0

December 2025minor

Added

Recursive context decomposition for complex queries
Multi-query batching for efficient bulk operations
Summary storage for LLM-generated document summaries
Semantic search with embedding-based similarity

Improved

Hybrid search combining keyword and semantic matching
Smart chunking for optimal section splitting
Relevance scoring with multi-factor ranking

Release 1

v1.0.0

November 2025major

Added

Initial release of Snipara Project Intelligence platform
MCP server integration for Claude Code, Cursor, and more
Core tools: snipara_context_query, snipara_search, snipara_inject, snipara_context
Token budgeting and metadata response
GitHub OAuth and email authentication
Team collaboration features
Stripe billing integration

Feedback

Want to request a feature or report a bug? Contact the Snipara team.