Akashic
Subscribe Open WorkspaceWorkspace

Private beta opens soon. Subscribe for Akashic product updates.

℗ 2026 Colossal Capital · managed SaaS

Any repo becomes a Collection.
Built for teams and every AI-built codebase.

Connect any Git repository — legacy monolith, greenfield service, Cursor project, Claude Fable, ChatGPT canvas, or Gemini app. Each repo is a Collection. Bundle Collections into a Library or Workspace, and iterate toward a shippable product — engineers, product, and identity-provisioned AI agents in one portal.

The first product in the Colossal suite — and the first to open itself to agents. Your Collection's corpus is available over MCP: Claude, ChatGPT, Cursor, and your own agents search the same reference corpus your tickets cite — authenticated, tenant-scoped, provenance-logged.

Open Workspace Repo → Collection → Ship

collection · library · workspace · human at the gate

AI-built repos · the timing

One-shot masterpieces need a label, not a landfill

Teams generate complete apps in single sessions. Akashic turns those spikes into maintained products.

What a one-shot build is

A Claude Fable, ChatGPT project, Gemini app, or similar one-shot AI build — a repo born from a single creative session. It runs, it impresses, and then it sits there because nobody knows what to do with it next.

What a Collection is

Any repository you connect — AI-built or not — becomes a Collection: analyzed, ticketed, and improved with a human at the gate.

What a Collection holds

Code, context, tickets, and docs — together

A Collection is the repo plus everything it takes to keep shipping: the code agents run against, how it was built, the tickets your team shares, and the docs they read.

Code

What agents run against

The connected repository — indexed so analysis, search, and agents work from the real tree, not a guess.

RepoSource, branches, and history
AnalysisGaps become tickets in an order a team can ship
SearchAnswers cite files, not model memory
Context

How it was made

Prompt history and playbooks stay with the Collection, so the next session starts where the last one left off.

PromptsHow the repo was actually built
PlaybooksConventions agents inherit
ReviewA human at the moments that matter
Tickets

Work the team can see

Analysis lands as tickets. Product and eng share one gate before anything syncs to your board.

BoardsLinear, Jira, GitHub Issues — after the gate
DedupeSame work is not filed twice
ApprovalPeople stay the approver
Docs

What the team reads

Documentation stays current as the Collection changes — the same Workspace engineers and product already use.

GuidesProduct and operator docs in one place
CoverageSee what is still missing
HistoryVersioned with the Collection

One-shot AI builds arrive heavy on prompts. Mature repos arrive heavy on tickets and docs. Same Collection, different balance.

Product structure

From repo to shippable product

Connect repos, bundle Collections, compose the Product Board, ship.

01 · REPO

Collection

Connect any Git repo. Analysis finds gaps, files tickets, and agents work them — with human approval at the gate.

One Collection per repo.
02 · BUNDLE

Library or Workspace

Add Collections to a Library (same entity — API, web, contracts, infra) or a Workspace (cross-entity — org, client, portfolio). Shared history and one approval gate.

The album — or the full roster in one room.
03 · COMPOSE

Product Board

Drag Collections into role slots — frontend, API, data, infra, docs — until the bundle is a shippable product.

Compose in the Workspace.
04 · SHIP

Reconcile & deploy

Tickets across the bundle, agent workflows, human approval, then ship under scoped identities.

Product and eng share the gate; agents run the work.

Workspace is where your team works Collections, composes the Product Board, reconciles tickets, and ships together. Every ticket, doc, and decision they produce is a Record — attributable and owned.

Product Board

Slot Collections into a shippable product

Analysis runs against the bundle; tickets file across repos.

Drag-and-drop composition

Each slot is a role in the product. One Collection per slot — clear ownership per lane.

App or infrastructure

Switch the board to emphasize product roles or the rails underneath — without mixing the two.

Ship from the bundle

Tickets and agent work run against the composed board — not a loose bag of repos.

Open the Product Board on any Library or Workspace in the Workspace.

Personas · the agentic bench

A pantheon, not a chatbot

Every task routes to a named persona — a role-shaped agent with its own tools, budget, and review posture — not a generalist model guessing at intent. The planner drafts the TRD; the editor writes the code; the reviewer, tester, and security personas gate it; the merger ships it. Each carries a mythic deity identity — Odin writes docs, Athena reports coverage, Hermes conducts — so the roster reads like what it is: a bench of specialists.

Roles, not prompts

Tickets route by content signal — a docs ticket lands on Odin (doc_writer), a security finding on Shangdi — so the right persona owns the right lane, every time.

The autonomous ensemble

Six in-process personas — planner, conductor, dispatcher, merger, indexer, analyst — run the campaign loop as a mailbox, not a monolith: each owns one link in the chain and reports to a shared blackboard.

Provisioned, not implied

Every persona registers in Idio's agent catalog before it can act — RBAC, budgets, and HITL gates are granted per seat, so a persona can't exceed its mandate.

See the live roster and per-persona activity on the Personas page.

Corpus · the shared substrate

Every answer cites evidence

Akashic's analysis doesn't guess — it routes your product against a curated corpus of exemplar code, docs, Stack Overflow evidence, and connectors across 145 domains. The same substrate is callable from outside: the Corpus MCP exposes routing, search, and citations to Claude Code, Cursor, ChatGPT, or any MCP client.

Route before you build

corpus_route answers "what should I build from" — the best exemplar, scored, with its license and alternates. Not a suggestion; a citation.

Evidence, not vibes

Every routed pick traces to real code or docs — the substrate is license-aware, so ingestible sources and reference-only services are never blurred.

It learns your outcomes

corpus_outcome reports whether a citation helped — and it calibrates routing to your tenant, not just the global model.

One MCP URL connects any client — anonymous browsing is rate-limited; an account key unlocks the sustained tier. Connection info →

Libraries

Multiple repos, one entity, one Library

Most real products span several repos. Group every related Collection into a Library and run analysis, tickets, and agents across the full stack.

collection/payments-apiindexed
collection/payments-webindexed
collection/shared-contractssyncing
collection/deploy-confignew

library · payments · 4 collections

Cross-collection view

See how Collections in the Library relate — before anyone merges.

Unified reconciliation

Tickets dedupe against Linear, Jira, and GitHub Issues across the whole Library. One gate.

Library-scoped agents

Agents per Collection with shared Library context — the whole album, not one track.

Collaborate · your Workspace

One place for eng, product, and agents

Akashic is a collaborative product — web portal, desktop app, and VS Code extension all sharing the same Workspace. Engineers run analyses and review agent PRs; product approves at the reconciliation gate.

Web portal

Analyses, chat with citations, ticket queue, agent approvals — no install.

Open your Workspace →

Desktop app

Run Akashic locally with a bundled local model — private and offline-capable by default. Switch to your own keys anytime; same Workspace, same gate.

Download the app →

VS Code extension

Connect Collections, run analysis, search with citations — same Workspace as the portal, from your editor.

VS Code Marketplace →

Corpus MCP

Plug Claude Code, Cursor, or ChatGPT into the shared evidence substrate — routing, exemplars, and citations over MCP. No portal account required to browse.

Connection info →

Chat with citations

Answers from indexed Collections — file paths, not model memory.

Managed CI/CD

Builds, scans, and deploys — routed to your targets.

Codebase, PM & documentation integrations

Connect where your code lives, where your team tracks work, and where your docs live — plus the stores and models behind the reasoning engine. OAuth through Idio — credentials stay in your vault; Akashic reads boards and docs for context and writes reconciled tickets back after the gate. Planned hosts are on the roadmap.

Code hosts

GitHub
GitLab
Bitbucket

Reasoning engine

Vector search
Knowledge graph
Model routing
Private transfer learning
Self-hosted embeddings
Code maps

PM tools

Jira
Linear
Asana
Monday.com
ClickUp
GitHub Issues
Bitbucket IssuesPlanned
Azure DevOpsPlanned
Google SheetsPlanned
Microsoft PlannerPlanned
Microsoft ListsPlanned

Documentation

Confluence
Notion
Linear
Monday.com
ClickUp
GitHub WikiPlanned
Google DocsPlanned
SharePointPlanned

Sandboxing

Daytona
E2B
Nightona
Firecracker

Agents never run code on your machines or your main checkout — every change is built, tested, and verified inside an isolated sandbox (Daytona, E2B, Nightona, or Firecracker microVMs). Only the reviewed diff leaves the sandbox.

Cloud providers

AWS
Google Cloud
Azure
DigitalOcean
OVHcloud
Oracle Cloud
IBM Cloud

Infrastructure as code

Terraform
OpenTofu
Pulumi
Terraspace

Pipelines & execution

GitHub Actions
GitLab CI/CD
Bitbucket Pipelines
Tekton
CircleCIPlanned
JenkinsPlanned

Deployments run through your pipeline of record — generated GitHub Actions, GitLab CI/CD, or Bitbucket Pipelines — on any major cloud. Enterprises can execute on Tekton inside their own cluster for per-run execution isolation. CircleCI and Jenkins configs are detected today; generation and run tracking are planned.

How work gets done · named identities

Agents execute. People approve.

Tickets that span the stack become ordered work for named specialists — planning, code, review, security, release, memory, and operations — each a scoped, revocable Idio identity, gated by a human where it matters. Odo, the platform helper, guides the flow; ObVect runs the workflows underneath.

Odo

Akashic — portal helper.

Guides onboarding and the other agents.

Platform agents — no first among equals.

👑

Apollo

Greek — leader of the Muses.

Orchestrates dispatch, HITL triage, and mandate alignment.

🏛

Sang-je

Korean — Heavenly Emperor.

Tenant sovereignty: connections, settings, and model-routing mandate.

◎

Marduk

Babylonian — Holder of the Tablets of Destiny.

Plans and roadmaps. Read-only on code.

☀

Olorun

Yoruba — owner of the sky.

Composes Product Boards. Read-only on code.

☽

Dyaus

Vedic — sky father of the open heaven.

Analysis, exploration, and product browsing. Read-only.

🕉

Shiva

Hindu — Lord of transformation.

Restructures the repo and consolidates docs before implementation begins.

⚡

Zeus

Greek — Olympian sky king.

Cloud and production deploys.

☀

Amaterasu

Shinto — celestial sun sovereign.

Frontend, UI, and experience.

🧵

Ixchel

Maya — weaver of life and medicine.

Backend and APIs.

⚒

Sunan Ambu

Sundanese — the noble mother, dweller of kahyangan.

Creates and tends the tools agents use.

📦

Waaq

Oromo — sky creator.

Releases, packages, and images.

⌂

Nyame

Akan — sky vault.

Data, schemas, and storage.

☘

Lugh

Celtic Irish — master of all crafts.

General code edits and PRs.

◉

Ra

Egyptian — sun who sees all.

Code review. No writes.

天

Shangdi

Chinese — Mandate of Heaven.

Security and compliance.

⬡

Heimdall

Norse — watchman of the sky-bridge.

Tests and coverage.

✎

Dewi Saraswati

Indonesian / Balinese — goddess of knowledge and flow.

Creates and syncs tickets across your boards.

✧

Odin

Norse — the rune-winner.

Docs and reports.

⚭

Mawu-Lisa

Fon — moon and sun made one.

Merges branches and verifies the closed loop.

🌊

Tangaroa

Māori — god of the ocean and its currents.

CI and pipelines.

☿

Mercury

Roman — swift messenger.

CDN, cache, rate limits, and network management.

☷

Pachamama

Andean — earth mother who holds what is built.

Deploy health and SLOs.

▣

Perun

Slavic — thunderer.

Alerts on breakage.

◇

Thought Woman

Pueblo / Tse Che Nako — she thinks the world into being.

Knowledge graph, codemap, and knowledge studio.

🌾

Mokosh

Slavic — Moist Mother Earth.

Multi-tenant, context-aware agent memory.

✉

Birrahgnooloo

Kamilaroi — messenger of Baiame.

Board mailbox — answers what agents are claiming, asking, and flagging in-flight.

〜

Quetzalcoatl

Aztec — feathered serpent of wind.

Skillfile creator — playbooks only, never opens code PRs.

✦

Kāne

Hawaiian — giver of life.

Private model adaptation and training.

𒀭

Anu

Mesopotamian — god of the firmament, the boundary between heaven and earth.

Boundary and access policy.

♛

Ishtar

Mesopotamian — queen of heaven.

Analytics and restore points.

☁

Tengri

Turkic — Eternal Blue Sky.

Cloud infrastructure health and cost.

⛈

Ukko

Finnish — sky thunderer.

Reasoning graphs.

🚪

Janus

Roman — two-faced gatekeeper.

Ticket reconciliation.

⚖

Maat

Egyptian — feather of truth.

UX simulation.

⚖

Ahura Mazda

Zoroastrian — Wise Lord of asha, truth against the lie.

Weighs work against contract and goals.

𓁟

Thoth

Egyptian — scribe of the gods.

Logging and portal interaction tracking.

𓆣

Seshat

Egyptian — goddess of accounting and measurement.

Beta application scoring.

Agent concurrency · by design

Unlimited configurations. Concurrent slots per tier.

Provision as many agents as you need — no per-person cap, no per-agent seat charge. Your Colossal Capital tier sets how many agents run at once. When slots are full, excess runs queue automatically and start when a running agent finishes. People stay the approver. Full concurrency & pricing →

Teams & organizations

Business 50 concurrent · Business+ 100 · Org 250 · Org+ 500 · Enterprise unlimited. Excess queues. Human seats are priced separately on team plans — agents are not sold per seat.

Individuals

Free 3 concurrent · Pro 10 · Ultra 25. Same rule: unlimited configurations, queue when slots are full. Ultra and above include a priority queue.

Governance · compliance

Every agent action — attributed, gated, and exportable

Autonomous, never unaccountable. Agents commit under their own provisioned identity, every action lands in a tenant-scoped audit trail, and the moments that matter are human-gated. Built for SOC 2 and GDPR.

Per-agent commit identity

Each commit, push, and PR is authored by that agent — clean Git blame, not an anonymous shared bot.

Full audit trail

Every agent action is recorded per tenant, filterable in the portal, and exportable on request.

Approval gates where they count

Release a ticket to an agent, approve a PR before merge, and sign off a production deploy — each optional per environment. Low-risk gates can auto-approve on calibrated confidence, so routine work never waits in a queue — everything else reaches a human.

Off-portal notifications

Every gate notifies a human with a deep link, so approvals never stall in a queue nobody's watching.

Enterprise auth

MFA, SSO, and user provisioning — through Idio.

GDPR-ready by default. The audit trail is downloadable and erasable on request — your data stays portable, and right-to-erasure is a button, not a support ticket. Observability and access policy are managed by ObVect →

Powered by ObVect

The engine behind the portal

Akashic is the product you use. ObVect is the AI backend underneath.

ObVect — the AI backend

Inference, retrieval, and agent runs — one API across the models you already use, grounded in your code and docs. Hosted keys, self-hosted GPU capacity on HuggingFace, Vast.ai, JarvisLabs, or Theta EdgeCloud, and custom OpenAI-compatible endpoints all route through the same gate. Supported providers →

IDENTITY · ENGINEERING · LEARNING

Named agents. Models that learn at your boundary.

Idio provisions every Akashic persona as a named, scoped, revocable identity. With consent, grants, and entitlement, an approved Collection or Record can inform a model for one user, one code/content resource, one entity, or an organization—without merging another entity's work.

USER

Personal

One engineer's approved workflows and preferences.

CONTENT

Collection or Record

One repository or artifact's authorized learning corpus.

ENTITY

One entity

Code intelligence and agents bounded to that entity.

ORGANIZATION

Across entities

Organization-wide models built only from approved corpora.

Private code and corpora do not roll up automatically. Shared ecosystem learning is separate, sanitized, and consent-controlled through Idio Data & Training.

Developed in Partnership with Ackward Roots Inc

Product newsletter

Updates on Akashic

Product news, rollout notes, and when beta access opens. Akashic itself is free — you pay for the ObVect tenant that runs your Workspace.

Product updates only. Unsubscribe anytime.