Local-first · AI memory · Agent runtime
Memory you can read.
Models you can pick.
Agents you can audit.
A local-first AI workspace. Your long-term memory lives as structured rows you can query — not opaque summaries. Pick the right model for each task: cloud frontier, local, or cheap. Every change the agent proposes lands in an approval queue first.
Desktop App
Download Kijulo
Run locally on your machine. Your data never leaves your environment.
Long-term memory
Memory you can read, edit, and query
Other AI tools remember by stashing opaque summaries you can never inspect. Here, the agent’s memory is your data: typed rows in a real database, every value provenance-tagged, every change approved before it lands.
Rows, not opaque summaries
Every fact the agent learns becomes a typed row you can query, edit, or delete. No black-box memory blob you can never inspect — your memory is data, in the open.
Real SQL access
Your memory lives in a real database. Query it with SQL — joins, aggregations, window functions. The agent reads it cheaply because it asks with WHERE clauses, not data dumps.
Provenance on every cell
See what came from you, what came from an agent, when, and with what context. Source Trace links every value back to the exact conversation that produced it.
Verified ground truth
Mark canonical data as Verified. Agents see a red warning before proposing any modification — and you decide row-by-row whether the flag survives.
Right model. Right task. Right cost.
Stop overpaying for simple tasks. Stop compromising on hard ones.
Locking every conversation to a single model is how AI bills run out of control. Kijulo lets you pick the model per conversation and per workflow — bring your own key, or use ours.
Cloud SOTA when accuracy matters
Drafting a critical document, reasoning over dense material? Pick the strongest frontier model from any major provider. Bring your own key.
Local model when privacy matters
Confidential, regulated, or contractually restricted data? Route the conversation to a local model — nothing leaves your machine. Full functionality, zero exposure.
Small model when neither does
Classifying a row, extracting a date, picking a category? Use a cheap, fast model. Save the frontier budget for the work that actually needs it.
Exact context, not data dumps
Your token bill stops growing with your data
Other AI tools dump your entire dataset into the context window. Kijulo passes the LLM exactly the rows you scoped it to — and lets the database do the heavy lifting.
You scope the context
Start a conversation from a row, a filtered view, a group, or nothing at all. The agent sees only what you scoped — not the table, not the database, not the history.
Flat cost curve
Add 10,000 rows tomorrow. Your token bill stays the same. Agents query with WHERE clauses; raw data never enters the context window.
Workflows compound savings
First run: the agent explores. Second run: it follows the proven playbook. Each replay costs less because the steps are already known.
Works without an LLM
Your schemas, views, and workflows function with no model attached. AI is an enhancement, not a dependency. If costs spike, you keep working.
Speed to velocity
A fast agent with no direction is expensive noise
Speed is how fast the agent runs. Velocity is how fast you get somewhere. Three primitives turn one into the other.
Workflows give direction
A chat that worked becomes a reusable playbook. The agent stops re-discovering and starts executing — same quality, every time, on any row you point it at.
Scheduling gives a beat
Run a workflow on a cron. Run it on new rows. Run it when a threshold is crossed. The agent works while you sleep, and you wake up to a review queue, not a chat log.
Pending Changes keep it honest
Every output the agent proposes lands in an approval queue. Review, edit, approve, reject — inline. The faster the agent runs, the more important this gate becomes.
How it works
From scattered data to running system
Define your data
Create entity types with typed, validated fields. Model your domain exactly the way you think about it. Your schema, your rules.
See it your way
Tables, boards, saved views, filters. The same data from any angle — no re-entry, no duplication.
Talk to your agents
Spawn conversations with the exact context you choose. Agents call tools, operate on your structured data, and record every step. Promote successful runs into reusable workflows.
Stay in control
Every change the agent proposes goes through an approval queue. Confidence scores, full provenance, one-click intervene. The agent works for you — not the other way around.
Why Kijulo
Built different
Local-first by default
Run on your machine. Your database, your files, your memory — all local. Go online when you choose to, not because you have to.
Transparency over magic
Every piece of data has provenance. Confidence scores, reasoning chains, full audit trail. Not a black box — a glass box.
Composable by design
Entity types are contracts. Templates transform. Workflows chain. Build simple pieces, then compose them into systems as your needs grow.
Built for control, not magic
No auto-execution. No silent overwrites. No hidden context. The agent proposes, you decide. Every time, on every value.
How we compare
A category of one — for now
Most AI tools fit into one of three buckets. None of them combine structured memory, model choice, and human-in-the-loop the way Kijulo does.
| Capability | Chat AI tools | AI note-taking apps | Cloud agent platforms | Kijulo |
|---|---|---|---|---|
| Long-term memory you can read and query | No | Limited | Partial | Yes |
| Run on a fully local model | No | No | No | Yes |
| Pick a different model per task | No | Limited | Limited | Yes |
| Human approval before changes land | N/A | No | Partial | Yes |
| Full provenance on every value | No | No | Partial | Yes |
| Workflows + scheduling | No | No | Yes | Yes |
| Your data file, on your disk | No | Partial | No | Yes |
Categories describe product types, not specific vendors. Capabilities vary by tier within each category.
Use cases
One platform. Your domain.
Define the structures that match your work. Use Kijulo as a pure data tool, or add agents to automate the parts you don’t want to do manually. The platform adapts to you.
Project Tracking
Define your own project structure — tasks, milestones, blockers, decisions. Track progress with boards and tables. Add dependencies, link related items, filter by status.
Decision Journal
Capture decisions with context, pros/cons, and outcomes. Link decisions to the situations that triggered them. Review past choices, spot patterns, build institutional memory.
Competitive Intelligence
Structure competitor data, track signals, synthesize insights. Let agents scan for updates, score signals by confidence, and surface what matters — you review and decide.
Content Production
Track ideas from capture through research to publication. Add agents to draft, research, or repurpose — with your voice encoded in templates.
FAQ
Frequently asked questions
Does Kijulo work offline?
Yes. The desktop app runs entirely on your machine. With a local LLM (Ollama or similar) attached, the full agent loop — conversations, workflows, scheduled runs, RAG search — works with no internet connection at all.
Can I use Kijulo without an LLM?
Yes. Your entity types, views, workflows, search, and data live and breathe with no model attached. The LLM is an enhancement layer, not a dependency. If your provider goes down or prices spike, your data and your tools keep working.
Where is my data stored?
In a local Postgres database on your machine, plus files in folders you explicitly grant access to. Nothing is uploaded by default. The Pro tier adds optional cloud sync across your own devices — you control whether to enable it.
Can I use my own LLM provider?
Yes. Bring your own API key for any major frontier provider, or point Kijulo at a local model running on your hardware. You pick the model per conversation and per workflow — no vendor lock-in.
Can the agent see all my files?
Only the folders you explicitly add to your Read Allowlist. Write access is a separate, stricter opt-in (the Write Allowlist) you grant directory by directory. The agent cannot touch anything outside those paths.
What is a Pending Change?
A proposed mutation from an agent — adding a row, editing a cell, writing a file — that waits in an approval queue until you review it. You see the diff, the confidence, and the originating conversation. Nothing lands until you approve.
Can I export my data?
Always. Your database is a real Postgres database with standard tooling. Export, back up, migrate, or run your own queries — no proprietary format, no lock-in.
How does the long-term memory differ from chat history?
Chat history is an opaque log nobody can query. Our memory is structured rows: typed fields, indexed, queryable, editable. Every agent run reads cheaply from the rows the previous runs left behind. The data table is the agent’s memory — and it is yours to inspect.
Pricing
Start free. Scale when you’re ready.
Try the web app at no cost. Upgrade to Standard for unlimited local power, or Pro for cloud sync and included LLM access.
Free
Web app at kijulo.com. Get started in seconds.
- Web app — no install needed
- 3 entity types, 100 rows
- 10 agent conversations/month
- All views (table, board, calendar, timeline)
- BYO LLM API key required for agent
- Community support
Standard
Desktop app (Tauri). Unlimited everything. Data on your machine.
- Desktop app — run locally, works offline
- Unlimited entity types, rows, agent conversations
- All views + all agent tools + workflows
- BYO LLM API key (cloud or local)
- Email support
Pro
Cloud sync + included LLM access. No API key needed.
- Everything in Standard
- Cloud sync — data across devices
- Included LLM access (500 requests/month)
- No API key needed
- Priority support
Enterprise
Shared databases, on-premise deployment, SSO, unlimited LLM.
- Everything in Pro
- Shared team databases
- SSO / SAML
- On-premise deployment
- Custom tool extensions
- Unlimited LLM access
- Dedicated support + SLA
How do you want to use Kijulo?
Same platform, different needs. Pick the path that fits you.
For Individuals
Make your AI work compound. Structure your data so every insight builds on the last. Run locally, own everything, and stop losing your best thinking to chat threads.
For Enterprise
Shared databases for your teams. On-premise deployment. Local LLMs. Custom tool extensions. Give every employee access to structured, AI-enhanced data — on your infrastructure, under your control.
Local-first by default
Human-in-the-loop
Transparency over magic
Your LLM, your choice
Enterprise ready
Your memory. Your models. Your machine.
Start with a local database. Define your structures. Add agents when you’re ready. Every step recorded, every output reviewed, every workflow repeatable.